Nothing
R/family.gait0.R
In VGAM: Vector Generalized Linear and Additive Models
Defines functions
y.gaitcombo.check
specialsvglm
moments.gaitdcombo.zeta
moments.gaitdcombo.log
moments.gaitdcombo.pois
moments.gaitdcombo.1par
gaitd.errorcheck
rgaitdbinom
qgaitdbinom
pgaitdbinom
dgaitdbinom
moments.gaitdcombo.binom
gaitdlog
gaitdzeta
Trunc
meangaitd
get.indices.gaitd
pd.damlm.old
pd.damlm
Documented in
dgaitdbinom
gaitd.errorcheck
gaitdlog
gaitdzeta
get.indices.gaitd
meangaitd
moments.gaitdcombo.1par
moments.gaitdcombo.binom
moments.gaitdcombo.log
moments.gaitdcombo.pois
moments.gaitdcombo.zeta
pd.damlm
pgaitdbinom
qgaitdbinom
rgaitdbinom
specialsvglm
Trunc
y.gaitcombo.check
# These functions are
# Copyright (C) 1998-2024 T.W. Yee, University of Auckland.
# All rights reserved.
amazon.col <- "#3b7a57"
avocado.col <- "#568203"
indigo.col <- "#6D5ACF" # Indigo http://hexcolor16.com/6d5acf
iris.col <- "#5a4fcf"
turquoise.col <- "#30d5c8" # for truncation
dirt.col <- "#9b7653"
deer.col <- "#ba8759"
desire.col <- "#ea3c53" # A dark red colour
peach.col <- "#ffe5b4"
azure.col <- "#007fff" # Unneeded
artichoke.col <- "#8f9779"
asparagus.col <- "#87a96b"
pd.damlm <-
function(ind.num, ind.den,
prob.num, prob.den = prob.num,
is.dipped = c(FALSE, FALSE)) {
if (is.dipped[1] && !is.dipped[2])
is.dipped <- c(FALSE, FALSE)
if (!any(is.dipped)) {
return(prob.num * ((ind.num == ind.den) - prob.den))
}
if (!is.dipped[1] && is.dipped[2]) # Negated.
return( prob.num * prob.den)
if (all(is.dipped)) # Negated.
return(-((ind.num == ind.den) - prob.num) * prob.den)
stop("am confused... should never reach here")
} # pd.damlm
if (FALSE)
pd.damlm.old <-
function(num.a = NULL, num.d = NULL,
den.a = NULL, den.d = NULL,
prob.num, prob.den = prob.num,
Denom = NA, eta.d.max = 0) {
if (length(num.d) && length(den.d)) {
if (num.a == num.d)
return((1 - prob.num) * (prob.num - exp(eta.d.max) / Denom)) else
return(-prob.num * (prob.den - exp(eta.d.max) / Denom))
}
if (length(num.a) && length(den.d)) {
return(-prob.num * (prob.den - exp(eta.d.max) / Denom))
}
if (length(num.a) && length(den.a)) {
if (num.a == num.d) return(prob.num * (1 - prob.num))
}
return(-prob.num * prob.den)
} # pd.damlm.old
get.indices.gaitd <- function(ind, indeta) {
ind.b <- indeta[ind, 'launch']
ind.b <- c(na.omit(ind.b))
ind.e <- indeta[ind, 'finish']
ind.e <- c(na.omit(ind.e))
ind.z <- NULL
if (length(ind.e) > 0)
for (jay in seq(length(ind.e)))
ind.z <- c(ind.z, seq(from = ind.b[jay], to = ind.e[jay]))
ind.z
} # get.ind.gaitd
meangaitd <-
function(theta.p,
fam = c("pois", "log", "zeta"), # "genpois0",
a.mix = NULL, i.mix = NULL, d.mix = NULL, #
a.mlm = NULL, i.mlm = NULL, d.mlm = NULL, #
truncate = NULL, max.support = Inf,
pobs.mix = 0, # scalar
pobs.mlm = 0, # vector of length a.mlm
pstr.mix = 0, # scalar
pstr.mlm = 0, # vector of length a.mlm
pdip.mix = 0, # scalar
pdip.mlm = 0, # vector of length d.mlm
byrow.aid = FALSE, # Applies to 'pobs.mlm' & 'pstr.mlm'
theta.a = theta.p, # scalar, else a 2-vector
theta.i = theta.p, # scalar, else a 2-vector
theta.d = theta.p, # scalar, else a 2-vector
...
) { # ... ignored currently.
fam.choices <- c("pois", "log", "zeta")
fam <- match.arg(fam[1], fam.choices)[1]
baseparams.argnames <-
switch(fam, "pois" = "lambda", "log" = "shape", "zeta" = "shape")
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
MM <- switch(fam, "pois" = 1, "log" = 1, "zeta" = 1)
if (MM != 1 && MM != 2)
stop("can only handle 1 or 2 parameters")
if (!length(a.mix)) pobs.mix <- 0 # Make sure for all
if (!length(a.mlm)) pobs.mlm <- 0
if (!length(i.mix)) pstr.mix <- 0
if (!length(i.mlm)) pstr.mlm <- 0
if (!length(d.mix)) pdip.mix <- 0
if (!length(d.mlm)) pdip.mlm <- 0
if (length(pobs.mix) != 1) stop("bad input for argument 'pobs.mix'")
if (length(pstr.mix) != 1) stop("bad input for argument 'pstr.mix'")
if (length(pdip.mix) != 1) stop("bad input for argument 'pdip.mix'")
if (length(a.mlm) && length(pobs.mlm) > length(a.mlm))
warning("bad input for argument 'pobs.mlm'?")
if (length(i.mlm) && length(pstr.mlm) > length(i.mlm))
warning("bad input for argument 'pstr.mlm'?")
if (length(d.mlm) && length(pdip.mlm) > length(d.mlm))
warning("bad input for argument 'pdip.mlm'?")
if (length(a.mlm))
pobs.mlm <- matrix(pobs.mlm, 1, # length(xx),
length(a.mlm), byrow = byrow.aid)
if (length(i.mlm))
pstr.mlm <- matrix(pstr.mlm, 1, # length(xx),
length(i.mlm), byrow = byrow.aid)
if (length(d.mlm))
pdip.mlm <- matrix(pdip.mlm, 1, # length(xx),
length(d.mlm), byrow = byrow.aid)
alist <- list( # x = xx, # theta.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid)
alist[[paste0(baseparams.argnames[1], ".p")]] <- theta.p[1]
alist[[paste0(baseparams.argnames[1], ".a")]] <- theta.a[1]
alist[[paste0(baseparams.argnames[1], ".i")]] <- theta.i[1]
alist[[paste0(baseparams.argnames[1], ".d")]] <- theta.d[1]
if (MM == 2) {
alist[[paste0(baseparams.argnames[2], ".p")]] <- theta.p[2]
alist[[paste0(baseparams.argnames[2], ".a")]] <- theta.a[2]
alist[[paste0(baseparams.argnames[2], ".i")]] <- theta.i[2]
alist[[paste0(baseparams.argnames[2], ".d")]] <- theta.d[2]
}
mlist <- alist
mlist$type.fitted <- "All"
mlist$moments2 <- TRUE
mom.fun <- paste0("moments.gaitdcombo.", fam)
Bits <- do.call(mom.fun, mlist)
if (length(c(a.mix, a.mlm, i.mix, i.mlm, d.mix, d.mlm))) {
Denom.p <- as.vector(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p <= 0))
stop("0s found in the denominator (variable 'Denom.p')")
Numer <- as.vector(1 -
(if (length(a.mix)) pobs.mix else 0) -
(if (length(i.mix)) pstr.mix else 0) +
(if (length(d.mix)) pdip.mix else 0) -
(if (length(a.mlm)) rowSums(rbind(pobs.mlm)) else 0) -
(if (length(i.mlm)) rowSums(rbind(pstr.mlm)) else 0) +
(if (length(d.mlm)) rowSums(rbind(pdip.mlm)) else 0))
if (!all(is.finite(Numer)))
warning("variable 'Numer' contains non-finite values")
if (min(Numer, na.rm = TRUE) < 0)
warning("variable 'Numer' has negative values")
} # Inflation or deflation
c(Bits$mean)
} # meangaitd
Trunc <- function(Range, mux = 2, location = 0, omits = TRUE) {
if (!is.finite(mux) || length(mux) != 1 || round(mux) != mux ||
mux < 1)
stop("argument 'mux' must be a positive integer")
if (any(!is.finite(Range)) || length(Range) < 2 ||
!all(round(Range) == Range))
stop("bad input in argument 'Range'")
if (length(Range) > 2)
Range <- range(Range, na.rm = TRUE) # For vglm() and na.omit()
Min <- Range[1]
Max <- Range[2]
allx <- location + (mux * Min):(mux * Max)
multiples <- location + mux * (Min:Max)
if (omits) {
ans <- setdiff(allx, multiples)
if (length(ans)) ans else NULL
} else {
multiples
}
} # Trunc
gaitdzeta <-
function(a.mix = NULL, i.mix = NULL,
d.mix = NULL,
a.mlm = NULL, i.mlm = NULL, # Unstructured probs are
d.mlm = NULL, # contiguous
truncate = NULL, max.support = Inf,
zero = c("pobs", "pstr", "pdip"), # Pruned, handles all 6
eq.ap = TRUE, eq.ip = TRUE, eq.dp = TRUE,
parallel.a = FALSE, parallel.i = FALSE,
parallel.d = FALSE,
lshape.p = "loglink",
lshape.a = lshape.p, # "logitlink", 20201117
lshape.i = lshape.p, # "logitlink", 20201117
lshape.d = lshape.p, # "logitlink", 20211011
type.fitted = c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"),
gshape.p = -expm1(-ppoints(7)),
gpstr.mix = ppoints(7) / 3, # ppoints(9) / 2,
gpstr.mlm = ppoints(7) / (3 + length(i.mlm)),
imethod = 1,
mux.init = c(0.75, 0.5, 0.75), # Order is A, I, D.
ishape.p = NULL, ishape.a = ishape.p,
ishape.i = ishape.p, ishape.d = ishape.p,
ipobs.mix = NULL, ipstr.mix = NULL, # 0.25,
ipdip.mix = NULL, # 0.01, # Easy but inflexible 0.01
ipobs.mlm = NULL, ipstr.mlm = NULL, # 0.25,
ipdip.mlm = NULL, # 0.01, # NULL, Easy but inflexible
byrow.aid = FALSE,
ishrinkage = 0.95,
probs.y = 0.35) {
mux.init <- rep_len(mux.init, 3)
if (length(a.mix) == 0) a.mix <- NULL
if (length(i.mix) == 0) i.mix <- NULL
if (length(d.mix) == 0) d.mix <- NULL
if (length(a.mlm) == 0) a.mlm <- NULL
if (length(i.mlm) == 0) i.mlm <- NULL
if (length(d.mlm) == 0) d.mlm <- NULL
if (length(truncate) == 0) truncate <- NULL
lowsup <- 1
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support,
min.support = lowsup)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltruncat <- length(truncate <- sort(truncate))
ltrunc.use <- ltruncat > 0 || !is.infinite(max.support)
if (is.finite(max.support))
stop("argument 'max.support' must be 'Inf'.")
lshape.p <- as.list(substitute(lshape.p))
eshape.p <- link2list(lshape.p)
lshape.p <- attr(eshape.p, "function.name")
lshape.p.save <- lshape.p
lpobs.mix <- "multilogitlink" # \omega_p
epobs.mix <- list() # zz NULL 4 now 20200907 coz 'multilogitlink'
eshape.a <- link2list(lshape.a)
lshape.a <- attr(eshape.a, "function.name")
lpstr.mix <- "multilogitlink" # \phi_p
epstr.mix <- list() # zz NULL 4 now 20200907 coz 'multilogitlink'
lpdip.mix <- "multilogitlink" # zz unsure 20211002
epdip.mix <- list() # zz unsure 20211002
eshape.i <- link2list(lshape.i)
lshape.i <- attr(eshape.i, "function.name")
eshape.d <- link2list(lshape.d)
lshape.d <- attr(eshape.d, "function.name")
lshape.p.save <- lshape.p
gshape.p.save <- gshape.p
if (is.vector(zero) && is.character(zero) && length(zero) == 3) {
if (li.mix + li.mlm == 0)
zero <- setdiff(zero, "pstr")
if (la.mix + la.mlm == 0)
zero <- setdiff(zero, "pobs")
if (ld.mix + ld.mlm == 0)
zero <- setdiff(zero, "pdip")
if (length(zero) == 0)
zero <- NULL # Better than character(0)
}
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
if (lall.len + ltruncat == 0 && is.infinite(max.support))
return(eval(substitute(
zetaff(lshape = .lshape.p.save ,
gshape = .gshape.p.save ,
zero = NULL),
list( .lshape.p.save = lshape.p.save,
.gshape.p.save = gshape.p.save ))))
if (!is.logical(eq.ap) || length(eq.ap) != 1)
stop("argument 'eq.ap' must be a single logical")
if (!is.logical(eq.ip) || length(eq.ip) != 1)
stop("argument 'eq.ip' must be a single logical")
if (!is.logical(parallel.a) || length(parallel.a) != 1)
stop("argument 'parallel.a' must be a single logical")
if (!is.logical(parallel.i) || length(parallel.i) != 1)
stop("argument 'parallel.i' must be a single logical")
if (!is.logical(parallel.d) || length(parallel.d) != 1)
stop("argument 'parallel.d' must be a single logical")
if (FALSE) { # Comment this out 2 allow default eq.ap=TRUE, etc.
if (la.mix <= 1 && eq.ap)
stop("<= one unstructured altered value (no 'shape.a')",
", so setting 'eq.ap = TRUE' is meaningless")
if (li.mix <= 1 && eq.ip)
stop("<= one unstructured inflated value (no 'shape.i')",
", so setting 'eq.ip = TRUE' is meaningless")
if (ld.mix <= 1 && eq.dp)
stop("<= one unstructured deflated value (no 'shape.d')",
", so setting 'eq.dp = TRUE' is meaningless")
if (la.mlm <= 1 && parallel.a) # Only \omega_1
stop("<= one altered mixture probability, 'pobs", a.mlm,
"', so setting 'parallel.a = TRUE' is meaningless")
if (li.mlm <= 1 && parallel.i) # Only \phi_1
stop("<= one inflated mixture probability, 'pstr", i.mlm,
"', so setting 'parallel.i = TRUE' is meaningless")
if (ld.mlm <= 1 && parallel.d) # Only \psi_1
stop("<= one deflated mixture probability, 'pdip", d.mlm,
"', so setting 'parallel.d = TRUE' is meaningless")
} # FALSE
type.fitted.choices <-
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s")
type.fitted <- match.arg(type.fitted[1], type.fitted.choices)[1]
tmp7a <- if (la.mlm) paste0("pobs.mlm", a.mlm) else NULL
tmp7b <- if (li.mlm) paste0("pstr.mlm", i.mlm) else NULL
tmp7c <- if (ld.mlm) paste0("pdip.mlm", d.mlm) else NULL
tmp3 <- c(shape.p = lshape.p,
pobs.mix = if (la.mix) "multilogitlink" else NULL,
shape.a = if (la.mix > 1) lshape.a else NULL,
pstr.mix = if (li.mix) "multilogitlink" else NULL,
shape.i = if (li.mix > 1) lshape.i else NULL,
pdip.mix = if (ld.mix) "multilogitlink" else NULL,
shape.d = if (ld.mix > 1) lshape.d else NULL,
if (la.mlm) rep("multilogitlink", la.mlm) else NULL,
if (li.mlm) rep("multilogitlink", li.mlm) else NULL,
if (ld.mlm) rep("multilogitlink", ld.mlm) else NULL)
Ltmp3 <- length(tmp3)
if (la.mlm + li.mlm + ld.mlm)
names(tmp3)[(Ltmp3 - la.mlm - li.mlm - ld.mlm + 1):Ltmp3] <-
c(tmp7a, tmp7b, tmp7c)
par1or2 <- 1 # 2
tmp3.TF <- c(TRUE, la.mix > 0, la.mix > 1,
li.mix > 0, li.mix > 1,
ld.mix > 0, ld.mix > 1,
la.mlm > 0, li.mlm > 0, ld.mlm > 0)
indeta.finish <- cumsum(c(par1or2, 1, par1or2,
1, par1or2,
1, par1or2,
la.mlm, li.mlm, ld.mlm,
ld.mlm + 1) * c(tmp3.TF, 1))
indeta.launch <- c(1, 1 + head(indeta.finish, -1))
indeta.launch <- head(indeta.launch, -1)
indeta.finish <- head(indeta.finish, -1)
indeta.launch[!tmp3.TF] <- NA # Not to be accessed
indeta.finish[!tmp3.TF] <- NA # Not to be accessed
indeta <- cbind(launch = indeta.launch,
finish = indeta.finish)
rownames(indeta) <- c("shape.p",
"pobs.mix", "shape.a",
"pstr.mix", "shape.i",
"pdip.mix", "shape.d",
"pobs.mlm", "pstr.mlm", "pdip.mlm")
M1 <- max(indeta, na.rm = TRUE)
predictors.names <- tmp3 # Passed into @infos and @initialize.
blurb1 <- "Z" # zz1
if (la.mlm + la.mix) blurb1 <- "Generally-altered z"
if (li.mlm + li.mix) blurb1 <- "Generally-inflated z"
if (ltrunc.use) blurb1 <- "Generally-truncated z"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && !ltrunc.use)
blurb1 <- "Generally-altered and -inflated z"
if ( (la.mlm + la.mix) && !(li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered and -truncated z"
if (!(la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-inflated and -truncated z"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered, -inflated and -truncated z"
if (ld.mlm + ld.mix) blurb1 <-
c(blurb1,
if (la.mlm + la.mix + li.mlm + li.mix) "and " else "Generally",
"-deflated ")
new("vglmff",
blurb = c(blurb1, "eta regression\n",
"(GAITD-zeta(shape.p)-",
"zeta(shape.a)-MLM-",
"zeta(shape.i)-MLM-\n",
"zeta(shape.d)-MLM generally)\n\n",
"Links: ",
namesof("shape.p", lshape.p, earg = eshape.p,
tag = FALSE),
if (la.mix > 0) c(", ", "multilogit(pobs.mix)"),
if (la.mix > 1) c(", ",
namesof("shape.a", lshape.a, eshape.a, tag = FALSE)),
if (la.mix && li.mix) ", \n ",
if (li.mix > 0) c( if (la.mix) "" else ", ",
"multilogit(pstr.mix)"),
if (li.mix > 1) c(", ",
namesof("shape.i", lshape.i, eshape.i, tag = FALSE)),
if (li.mix && ld.mix) ", \n ",
if (ld.mix > 0) c( if (li.mix) "" else ", ",
"multilogit(pdip.mix)"),
if (ld.mix > 1) c(", ",
namesof("shape.d", lshape.d, eshape.d, tag = FALSE)),
if (la.mlm) paste0(",\n",
paste0(" multilogit(", tmp7a, collapse = "),\n"),
")") else NULL,
if (li.mlm) paste0(",\n",
paste0(" multilogit(", tmp7b, collapse = "),\n"),
")") else NULL,
if (ld.mlm) paste0(",\n",
paste0(" multilogit(", tmp7c, collapse = "),\n"),
")") else NULL),
constraints = eval(substitute(expression({
M1 <- max(extra$indeta, na.rm = TRUE)
la.mix <- ( .la.mix )
li.mix <- ( .li.mix )
ld.mix <- ( .ld.mix )
la.mlm <- ( .la.mlm )
li.mlm <- ( .li.mlm )
ld.mlm <- ( .ld.mlm )
use.mat.mlm.a <- if (la.mlm) {
if ( .parallel.a ) matrix(1, la.mlm, 1) else diag(la.mlm)
} else {
NULL
}
use.mat.mlm.i <- if (li.mlm) {
if ( .parallel.i ) matrix(1, li.mlm, 1) else diag(li.mlm)
} else {
NULL
}
use.mat.mlm.d <- if (ld.mlm) {
if ( .parallel.d ) matrix(1, ld.mlm, 1) else diag(ld.mlm)
} else {
NULL
}
if (la.mlm + li.mlm + ld.mlm == 0) {
Use.mat <- use.mat.mlm <- cbind(M) # shape.p only
}
if (la.mlm + li.mlm + ld.mlm) {
nc1 <- if (length(use.mat.mlm.a)) ncol(use.mat.mlm.a) else 0
nc2 <- if (length(use.mat.mlm.i)) ncol(use.mat.mlm.i) else 0
nc3 <- if (length(use.mat.mlm.d)) ncol(use.mat.mlm.d) else 0
use.mat.mlm <- cbind(1, matrix(0, 1, nc1 + nc2 + nc3))
if (la.mlm)
use.mat.mlm <- rbind(use.mat.mlm,
cbind(matrix(0, la.mlm, 1),
use.mat.mlm.a,
if (length(use.mat.mlm.i) == 0)
NULL else matrix(0, la.mlm, nc2),
if (length(use.mat.mlm.d) == 0)
NULL else matrix(0, la.mlm, nc3)))
if (li.mlm )
use.mat.mlm <-
rbind(use.mat.mlm,
cbind(matrix(0, li.mlm, 1 + nc1),
use.mat.mlm.i,
matrix(0, li.mlm, nc3)))
if (ld.mlm)
use.mat.mlm <-
rbind(use.mat.mlm, # zz1 next line:
cbind(matrix(0, ld.mlm, 1 + nc1 + nc2),
use.mat.mlm.d))
} # la.mlm + li.mlm
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
use.mat.mix <- cm3gaitd( .eq.ap , .eq.ip , .eq.dp , npar = 1)
tmp3.subset <- tmp3.TF[-(8:10)]
use.mat.mix <- use.mat.mix[tmp3.subset, , drop = FALSE]
notall0 <- function(x) !all(x == 0)
use.mat.mix <- use.mat.mix[, apply(use.mat.mix, 2, notall0),
drop = FALSE]
if (la.mix + li.mix + ld.mix > 0)
Use.mat <- use.mat.mix
if (la.mlm + li.mlm + ld.mlm > 0) {
Use.mat <- rbind(use.mat.mix,
matrix(0, nrow(use.mat.mlm) - 1, # bottom
ncol(use.mat.mix)))
Use.mat <- cbind(Use.mat,
matrix(0, nrow(Use.mat), # RHS
ncol(use.mat.mlm) - 1))
Use.mat[row(Use.mat) > nrow(use.mat.mix) &
col(Use.mat) > ncol(use.mat.mix)] <-
use.mat.mlm[-1, -1]
} # la.mlm + li.mlm + ld.mlm > 0
if (is.null(constraints)) {
constraints <-
cm.VGAM(Use.mat, x = x, apply.int = TRUE, # FALSE
bool = .eq.ap || .eq.ip || .eq.dp ||
.parallel.a || .parallel.i || .parallel.d ,
constraints = constraints) # FALSE
} # is.null(constraints)
if (la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm)
constraints <-
cm.zero.VGAM(constraints, x = x, .zero , M = M, M1 = M1,
predictors.names = paste0(predictors.names,
names(predictors.names)))
}),
list( .zero = zero, .tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.eq.ap = eq.ap, .eq.ip = eq.ip, .eq.dp = eq.dp,
.parallel.a = parallel.a, .parallel.i = parallel.i,
.parallel.d = parallel.d,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix ))),
infos = eval(substitute(function(...) {
list(M1 = .M1 ,
Q1 = 1,
dpqrfun = "gaitdzeta",
link = .predictors.names , # ...strips... from above
link1parameter = as.logical( .lall.len <= 2), # <= 1 safer
mixture.links = any(c( .la.mlm , .li.mlm , .ld.mlm ,
.la.mix , .li.mix ,
.ld.mix ) > 1), # FALSE if NULL
a.mix = as.vector( .a.mix ), # Handles NULL
a.mlm = as.vector( .a.mlm ),
i.mix = as.vector( .i.mix ),
i.mlm = as.vector( .i.mlm ),
d.mix = as.vector( .d.mix ),
d.mlm = as.vector( .d.mlm ),
truncate = as.vector( .truncate ),
max.support = as.vector( .max.support ),
Support = c( .lowsup , Inf, 1), # a(b)c format as a,c,b.
expected = TRUE,
multipleResponses = FALSE, # zetaff can be called if TRUE
parameters.names = names( .predictors.names ),
parent.name = c("zetaff", "zeta"),
type.fitted = as.vector( .type.fitted ),
type.fitted.choices = ( .type.fitted.choices ),
baseparams.argnames = "shape",
MM1 = 1, # One parameter for 1 response (shape). Needed.
zero = .zero )
}, list( .zero = zero, .lowsup = lowsup,
.type.fitted = type.fitted,
.type.fitted.choices = type.fitted.choices,
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix,
.truncate = truncate, .max.support = max.support,
.predictors.names = predictors.names,
.M1 = M1, .lall.len = lall.len
))),
rqresslot = eval(substitute(
function(mu, y, w, eta, extra = NULL) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least 1 shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vecs
ind.shape.z <- c(na.omit(ind.shape.z)) # At least 1 value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
scrambleseed <- runif(1) # To scramble the seed
qnorm(runif(length(y),
pgaitdzeta(y - 1, shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm),
pgaitdzeta(y , shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)))
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
initialize = eval(substitute(expression({
extra$indeta <- ( .indeta ) # Avoids recomputing it
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix +
la.mlm + li.mlm + ld.mlm
truncate <- as.vector( .truncate )
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(y)
M <- NOS * M1
tmp3.TF <- ( .tmp3.TF )
temp5 <-
w.y.check(w = w, y = y,
Is.nonnegative.y = TRUE,
Is.integer.y = TRUE,
ncol.w.max = 1, # Since max.support = 9 is possible
ncol.y.max = 1,
out.wy = TRUE, colsyperw = 1, maximize = TRUE)
w <- temp5$w
y <- temp5$y
glist <- y.gaitcombo.check(y, truncate = truncate,
a.mlm = a.mlm, a.mix = a.mix,
i.mlm = i.mlm, i.mix = i.mix,
d.mlm = d.mlm, d.mix = d.mix,
max.support = .max.support ,
min.support = .min.support )
extra$skip.mix.a <- glist$skip.mix.a
extra$skip.mix.i <- glist$skip.mix.i
extra$skip.mix.d <- glist$skip.mix.d
extra$skip.mlm.a <- glist$skip.mlm.a
extra$skip.mlm.i <- glist$skip.mlm.i
extra$skip.mlm.d <- glist$skip.mlm.d
extra$NOS <- NOS <- ncoly <- ncol(y) # Number of species
extra$type.fitted <- as.vector( .type.fitted )
extra$mux.init <- as.vector( .mux.init )
extra$colnames.y <- colnames(y)
extra$M1 <- M1
extra$index.M <- iam(NA, NA, M, both = TRUE) # Used in @weight
predictors.names <- ( .predictors.names ) # Got it, named
if (!length(etastart)) {
shape.p.init <- if (length( .ishape.p )) .ishape.p else {
zetaff.Loglikfun <- function(shapeval, y, x, w, extraargs) {
sum(c(w) * dzeta(x = y, shape = shapeval, log = TRUE))
}
shape.p.grid <- ( .gshape.p )
grid.search(shape.p.grid, objfun = zetaff.Loglikfun,
y = y, w = w)
}
shape.p.init <- rep(shape.p.init, length = n)
shape.d.init <-
shape.a.init <- shape.i.init <- shape.p.init # Needed
etastart <- matrix(nrow = n, ncol = M,
theta2eta(shape.p.init, .lshape.p , earg = .eshape.p ))
mux.more.a <- extra$mux.init[1] # 0.75 Err 2 slightly smaller
init.pobs.mix <- numeric(n)
if (tmp3.TF[ 2]) { # la.mix > 0
init.pobs.mix <- if (length( .ipobs.mix )) {
rep_len( .ipobs.mix , n)
} else {
is.a.mix1 <- rowSums(extra$skip.mix.a) > 0
rep(mux.more.a * sum(w[is.a.mix1]) / sum(w), n)
}
} # la.mix > 0
if (tmp3.TF[ 3]) { # Assign coln 3; la.mix > 1
shape.a.init <- if (length( .ishape.a ))
rep_len( .ishape.a , n) else shape.p.init # A vector
etastart[, 3] <-
theta2eta(shape.a.init, .lshape.a , earg = .eshape.a )
}
init.pstr.mix <- init.pdip.mix <- numeric(n)
try.gridsearch.pstr.mix <- FALSE
if (tmp3.TF[ 4]) { # li.mix > 0
init.pstr.mix <- if (length( .ipstr.mix )) {
rep_len( .ipstr.mix , n)
} else {
try.gridsearch.pstr.mix <- TRUE
numeric(n) # Overwritten by gridsearch
}
} # li.mix > 0
if (tmp3.TF[ 5]) { # li.mix > 1
shape.i.init <- if (length( .ishape.i ))
rep_len( .ishape.i , n) else shape.p.init # A vector
etastart[, (extra$indeta[5, 'launch'])] <-
theta2eta(shape.i.init, .lshape.i , earg = .eshape.i )
} # li.mix > 1
if (tmp3.TF[ 8]) { # la.mlm
init.pobs.mlm <- if (length( .ipobs.mlm )) {
matrix( .ipobs.mlm , n, la.mlm, byrow = .byrow.aid )
} else {
mux.more.a <- extra$mux.init[1]
init.pobs.mlm <- colSums(c(w) *
extra$skip.mlm.a) / colSums(w)
init.pobs.mlm <- init.pobs.mlm * as.vector( mux.more.a )
matrix(init.pobs.mlm, n, la.mlm, byrow = TRUE)
}
} else {
init.pobs.mlm <- matrix(0, n, 1)
}
try.gridsearch.pstr.mlm <- FALSE
if (tmp3.TF[ 9]) { # li.mlm
try.gridsearch.pstr.mlm <- !(length( .ipstr.mlm ))
init.pstr.mlm <- 0 # Might be overwritten by gridsearch
if (length( .ipstr.mlm ))
init.pstr.mlm <- as.vector( .ipstr.mlm )
init.pstr.mlm <- matrix(init.pstr.mlm, n, li.mlm,
byrow = .byrow.aid )
} else {
init.pstr.mlm <- matrix(0, n, 1)
}
init.pdip.mlm <- matrix(0, n, 2) # rowSums() needs > 1 colns.
gaitdzeta.Loglikfun1.mix <-
function(pstr.mix.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mix = pstr.mix.val,
pstr.mlm = extraargs$pstr.mlm, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdzeta.Loglikfun1.mlm <-
function(pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mlm = pstr.mlm.val,
pstr.mix = extraargs$pstr.mix, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdzeta.Loglikfun2 <-
function(pstr.mix.val, pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mix = pstr.mix.val,
pstr.mlm = pstr.mlm.val,
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
if (li.mix + li.mlm) {
extraargs <- list(
shape.p = shape.p.init,
shape.a = shape.a.init,
shape.i = shape.i.init,
shape.d = shape.d.init,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
pobs.mix = init.pobs.mix ,
pobs.mlm = init.pobs.mlm ,
pdip.mix = init.pdip.mix ,
pdip.mlm = init.pdip.mlm )
pre.warn <- options()$warn
options(warn = -1) # Ignore warnings during gridsearch
try.this <-
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
grid.search2( .gpstr.mix , .gpstr.mlm ,
objfun = gaitdzeta.Loglikfun2,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mix) {
extraargs$pstr.mlm <- init.pstr.mlm
grid.search ( .gpstr.mix ,
objfun = gaitdzeta.Loglikfun1.mix,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mlm) {
extraargs$pstr.mix <- init.pstr.mix
grid.search ( .gpstr.mlm ,
objfun = gaitdzeta.Loglikfun1.mlm,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
}
options(warn = pre.warn) # Restore warnings
if (any(is.na(try.this)))
warning("gridsearch returned NAs. It's going to crash.",
immediate. = TRUE)
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
init.pstr.mix <- rep_len(try.this["Value1"], n)
init.pstr.mlm <- matrix(try.this["Value2"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'",
" and/or 'gpstr.mlm = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mix) {
init.pstr.mix <- rep_len(try.this["Value"], n)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mlm) {
init.pstr.mlm <- matrix(try.this["Value"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mlm = seq(5) / 100'.")
}
} # la.mix + lnf.mix
mux.more.d <- extra$mux.init[3]
if (ld.mix) {
init.pdip.mix <- if (length( .ipdip.mix ))
rep_len( .ipdip.mix, n) else {
is.d.mix1 <- rowSums(extra$skip.mix.d) > 0
rep(mux.more.d * sum(w[is.d.mix1]) / sum(w), n)
}
} # ld.mix
if (ld.mlm) {
init.pdip.mlm <- if (length( .ipdip.mlm ))
matrix( .ipdip.mlm, n, ld.mlm, byrow = TRUE) else {
is.d.mlm1 <- rowSums(extra$skip.mlm.d) > 0
matrix(mux.more.d * (sum(w[is.d.mlm1]) / sum(w)) / ld.mlm,
n, ld.mlm)
}
} # ld.mlm
while (any((vecTF <- init.pobs.mix + init.pstr.mix + # -
init.pdip.mix +
rowSums(init.pobs.mlm) +
rowSums(init.pstr.mlm) + # -
rowSums(init.pdip.mlm) > 0.96875))) {
init.pobs.mix[vecTF] <- 0.875 * init.pobs.mix[vecTF]
init.pstr.mix[vecTF] <- 0.875 * init.pstr.mix[vecTF]
init.pdip.mix[vecTF] <- 0.875 * init.pdip.mix[vecTF]
init.pobs.mlm[vecTF, ] <- 0.875 * init.pobs.mlm[vecTF, ]
init.pstr.mlm[vecTF, ] <- 0.875 * init.pstr.mlm[vecTF, ]
init.pdip.mlm[vecTF, ] <- 0.875 * init.pdip.mlm[vecTF, ]
} # while
Numer.init1 <- 1 - rowSums(init.pobs.mlm) -
rowSums(init.pstr.mlm) - # +
rowSums(init.pdip.mlm) -
init.pobs.mix - init.pstr.mix - # +
init.pdip.mix # Differs from 'Numer'.
etastart.z <- if (lall.len == 0) NULL else {
tmp.mat <- cbind(if (tmp3.TF[ 2]) init.pobs.mix else NULL,
if (tmp3.TF[ 4]) init.pstr.mix else NULL,
if (tmp3.TF[ 6]) init.pdip.mix else NULL,
if (tmp3.TF[ 8]) init.pobs.mlm else NULL,
if (tmp3.TF[ 9]) init.pstr.mlm else NULL,
if (tmp3.TF[10]) init.pdip.mlm else NULL,
Numer.init1)
multilogitlink(tmp.mat)
} # etastart.z
if (!is.matrix(etastart.z)) etastart.z <- cbind(etastart.z)
nextone <- 1 # Might not be used actually
if (tmp3.TF[ 2]) {
etastart[, 2] <- etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 4]) { # Coln 2 or 4
etastart[, (extra$indeta[4, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 6]) { # Coln 2 or 4 or 6
etastart[, (extra$indeta[6, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 8]) {
ind8 <- (extra$indeta[8, 'launch']):(
extra$indeta[8, 'finish'])
etastart[, ind8] <- etastart.z[, nextone:(nextone +
la.mlm - 1)]
nextone <- nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (extra$indeta[9, 'launch']):(
extra$indeta[9, 'finish'])
etastart[, ind9] <- etastart.z[, nextone:(
nextone + li.mlm - 1)]
nextone <- nextone + li.mlm
}
if (tmp3.TF[10]) {
ind0 <- (extra$indeta[10, 'launch']):(
extra$indeta[10, 'finish'])
etastart[, ind0] <- etastart.z[, nextone:(nextone +
ld.mlm - 1)]
if (ncol(etastart.z) != nextone + ld.mlm - 1)
stop("miscalculation")
}
}
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.ishape.p = ishape.p,
.ishape.a = ishape.a,
.ishape.i = ishape.i,
.ishape.d = ishape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.lpdip.mix = lpdip.mix,
.epdip.mix = epdip.mix,
.ipstr.mix = ipstr.mix, .ipobs.mix = ipobs.mix,
.ipstr.mlm = ipstr.mlm, .ipobs.mlm = ipobs.mlm,
.ipdip.mix = ipdip.mix,
.ipdip.mlm = ipdip.mlm,
.byrow.aid = byrow.aid,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support,
.min.support = lowsup,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.predictors.names = predictors.names,
.mux.init = mux.init,
.gshape.p = gshape.p,
.gpstr.mix = gpstr.mix, # .gpdip.mix = gpdip.mix,
.gpstr.mlm = gpstr.mlm, # .gpdip.mlm = gpdip.mlm,
.ishrinkage = ishrinkage, .probs.y = probs.y,
.indeta = indeta,
.imethod = imethod, .type.fitted = type.fitted ))),
linkinv = eval(substitute(function(eta, extra = NULL) {
n.obs <- NROW(eta)
type.fitted <-
if (length(extra$type.fitted)) extra$type.fitted else {
warning("cannot find 'type.fitted'. Returning the 'mean'.")
"mean"
}
type.fitted <-
match.arg(type.fitted[1],
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"))[1]
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
morework <- type.fitted != "mean" # For efficiency
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(eta) / M1
Bits <- moments.gaitdcombo.zeta(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
if (morework) {
Denom.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p == 0)) {
smallval <- min(Denom.p[Denom.p > 0])
Denom.p[Denom.p == 0] <- 1e-09 # smallval
warning("0s found in variable 'Denom.p'. Trying to fix it.")
}
Numer <- c(1 - pobs.mix - pstr.mix + pdip.mix -
(if (la.mlm) rowSums(pobs.mlm) else 0) -
(if (li.mlm) rowSums(pstr.mlm) else 0) +
(if (ld.mlm) rowSums(pdip.mlm) else 0))
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom.p)
} # morework
if (!la.mlm && type.fitted %in% c("pobs.mlm")) {
warning("No altered MLM values; returning an NA")
return(NA)
}
if (!li.mlm && type.fitted %in% c("sum.mlm.i", "pstr.mlm")) {
warning("No inflated MLM values; returning an NA")
return(NA)
}
if (!ld.mlm && type.fitted %in% c("sum.mlm.d", "pdip.mlm")) {
warning("No deflated MLM values; returning an NA")
return(NA)
}
if (!la.mix && type.fitted %in% c("Pobs.mix")) {
warning("No altered mixture values; returning an NA")
return(NA)
}
if (!li.mix && type.fitted %in% c("sum.mix.i", "Pstr.mix")) {
warning("No inflated mixture values; returning an NA")
return(NA)
}
if (!ld.mix && type.fitted %in% c("sum.mix.d", "Pdip.mix")) {
warning("No deflated mixture values; returning an NA")
return(NA)
}
if (la.mix && morework) {
tmp13 <- # dpois() does not retain the matrix format
dzeta(matrix(a.mix, n.obs, la.mix, byrow = TRUE),
matrix(shape.a, n.obs, la.mix)) / (
c(Bits[["SumA0.mix.a"]]))
dim(tmp13) <- c(n.obs, la.mix)
dimnames(tmp13) <- list(rownames(eta),
as.character(a.mix))
propn.mat.a <- tmp13
} # la.mix
if (li.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dzeta(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.i, n.obs, li.mix)) / (
c(Bits[["SumI0.mix.i"]]))
dim(tmp55) <- c(n.obs, li.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(i.mix))
propn.mat.i <- tmp55 # Correct dimension
} # li.mix
if (ld.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dzeta(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.d, n.obs, ld.mix)) / (
c(Bits[["SumD0.mix.d"]]))
dim(tmp55) <- c(n.obs, ld.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(d.mix))
propn.mat.d <- tmp55 # Correct dimension
} # ld.mix
ans <- switch(type.fitted,
"mean" = Bits[["mean"]], # Unconditional mean
"shapes" = cbind(shape.p,
if (tmp3.TF[ 3]) shape.a else NULL,
if (tmp3.TF[ 5]) shape.i else NULL,
if (tmp3.TF[ 7]) shape.d else NULL),
"pobs.mlm" = pobs.mlm, # aka omegamat, n x la.mlm
"pstr.mlm" = pstr.mlm, # aka phimat, n x li.mlm
"pdip.mlm" = pdip.mlm, # aka psimat, n x ld.mlm
"pobs.mix" = pobs.mix, # n-vector
"pstr.mix" = pstr.mix, # n-vector
"pdip.mix" = pdip.mix, # n-vector
"Pobs.mix" = c(pobs.mix) * propn.mat.a, # matrix
"Pstr.mix" = c(pstr.mix) * propn.mat.i,
"Pdip.mix" = c(pdip.mix) * propn.mat.d,
"nonspecial" = probns,
"Numer" = Numer,
"Denom.p" = Denom.p,
"sum.mlm.i" = pstr.mlm + Numer *
dzeta(matrix(i.mlm, n.obs, li.mlm, byrow = TRUE),
matrix(shape.p, n.obs, li.mlm)) / Denom.p,
"sum.mlm.d" = -pdip.mlm + Numer *
dzeta(matrix(d.mlm, n.obs, ld.mlm, byrow = TRUE),
matrix(shape.p, n.obs, ld.mlm)) / Denom.p,
"sum.mix.i" = c(pstr.mix) * propn.mat.i + Numer *
dzeta(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.p, n.obs, li.mix)) / Denom.p,
"sum.mix.d" = -c(pdip.mix) * propn.mat.d + Numer *
dzeta(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.p, n.obs, ld.mix)) / Denom.p,
"ptrunc.p" = Bits[["SumT0.p"]] + 1 - Bits[["cdf.max.s"]],
"cdf.max.s" = Bits[["cdf.max.s"]]) # Pr(y <= max.support)
ynames.pobs.mlm <- as.character(a.mlm) # Works with NULLs
ynames.pstr.mlm <- as.character(i.mlm) # Works with NULLs
ynames.pdip.mlm <- as.character(d.mlm) # Works with NULLs
if (length(ans))
label.cols.y(ans, NOS = NOS, colnames.y =
switch(type.fitted,
"shapes" = c("shape.p", "shape.a", # Some colns NA
"shape.i", "shape.d")[(tmp3.TF[c(1, 3, 5, 7)])],
"Pobs.mix" = as.character(a.mix),
"sum.mix.i" = , #
"Pstr.mix" = as.character(i.mix),
"sum.mix.d" = , #
"Pdip.mix" = as.character(d.mix),
"pobs.mlm" = ynames.pobs.mlm,
"sum.mlm.i" = , #
"pstr.mlm" = ynames.pstr.mlm,
"sum.mlm.d" = , #
"pdip.mlm" = ynames.pdip.mlm,
extra$colnames.y)) else ans
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
last = eval(substitute(expression({
pred.names <- c( .predictors.names ) # Save it
link.names <- as.vector( .predictors.names )
parameter.names <- names(pred.names)
predictors.names <- NULL
for (jay in seq(M))
predictors.names <- c(predictors.names,
namesof(parameter.names[jay], link.names[jay], tag = FALSE,
earg = list())) # This isnt perfect; info is lost
misc$predictors.names <- predictors.names # Useful for coef()
misc$link <- link.names #
names(misc$link) <- parameter.names #
misc$earg <- vector("list", M1)
names(misc$earg) <- names(misc$link)
misc$earg[[1]] <- ( .eshape.p ) # First one always there
iptr <- 1
if (tmp3.TF[ 2])
misc$earg[[(iptr <- iptr + 1)]] <- list() # multilogitlink
if (tmp3.TF[ 3])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.a )
if (tmp3.TF[ 4])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 5])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.i )
if (tmp3.TF[ 6])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 7])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.d )
if (tmp3.TF[ 8]) { # la.mlm
for (ii in seq(la.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # la.mlm
if (tmp3.TF[ 9]) { # li.mlm
for (ii in seq(li.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # li.mlm
if (tmp3.TF[10]) { # ld.mlm
for (ii in seq(ld.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # ld.mlm
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.predictors.names = predictors.names,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
loglikelihood = eval(substitute(
function(mu, y, w, residuals = FALSE, eta,
extra = NULL, summation = TRUE) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
if (residuals) {
stop("loglikelihood residuals not implemented yet")
} else {
ll.elts <- c(w) *
dgaitdzeta(y, shape.p, log = TRUE, # byrow.aid = F,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)
if (summation) {
sum(ll.elts)
} else {
ll.elts
}
}
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
vfamily = c("gaitdzeta"),
validparams = eval(substitute(function(eta, y, extra = NULL) {
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
small. <- 1e-14
pobs.mix <- pstr.mix <- pdip.mix <- small. # 4 rowSums():
pobs.mlm <- pstr.mlm <- pdip.mlm <- matrix(small., NROW(eta), 1)
shape.a <- shape.i <- shape.d <- 0.5 # Needed
if (!is.matrix(eta)) eta <- as.matrix(eta)
shape.p <-
cbind(eta2theta(eta[, 1], .lshape.p , earg = .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 1] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
okay.mlm <-
all(is.finite(pobs.mlm)) && all(0 < pobs.mlm) &&
all(is.finite(pstr.mlm)) && all(0 < pstr.mlm) &&
all(is.finite(pdip.mlm)) && all(0 < pdip.mlm)
okay.mix <-
all(is.finite(shape.p)) && all(0 < shape.p) &&
all(is.finite(shape.a)) && all(0 < shape.a) &&
all(is.finite(shape.i)) && all(0 < shape.i) &&
all(is.finite(shape.d)) && all(0 < shape.d) &&
all(is.finite(pobs.mix)) && all(0 < pobs.mix) &&
all(is.finite(pstr.mix)) && all(0 < pstr.mix) &&
all(is.finite(pdip.mix)) && all(0 < pdip.mix) &&
all(pobs.mix + pstr.mix + pdip.mix +
rowSums(pobs.mlm) + rowSums(pstr.mlm) +
rowSums(pdip.mlm) < 1) # Combined
okay.mlm && okay.mix
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
simslot = eval(substitute(
function(object, nsim) {
pwts <- if (length(pwts <- object@prior.weights) > 0)
pwts else weights(object, type = "prior")
if (any(pwts != 1))
warning("ignoring prior weights")
eta <- predict(object)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
extra <- object@extra
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF )
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A AMLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
rgaitdzeta(nsim * length(shape.p), shape.p,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = .truncate , max.support = .max.support )
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
deriv = eval(substitute(expression({
tmp3.TF <- ( .tmp3.TF )
calA.p <- tmp3.TF[ 2]
calI.p <- tmp3.TF[ 4]
calD.p <- tmp3.TF[ 6]
calA.np <- tmp3.TF[ 8]
calI.np <- tmp3.TF[ 9]
calD.np <- tmp3.TF[10]
Denom1.a <- Denom1.i <- Denom1.d <-
Denom2.i <- Denom2.d <- 0 # Denom2.a is unneeded
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted.
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
minprob.baseline <- min(allprobs[, ncol(allprobs)], na.rm = TRUE)
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1) {
warning("fitted probabilities numerically 0 or 1 occurred")
} else
if (minprob.baseline < 0.10)
warning("Minimum baseline (reserve) probability close to 0")
if (control$trace)
cat("Minimum baseline (reserve) probability = ",
format(minprob.baseline, digits = 3), "\n")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- ncol(eta) / M1 # extra$NOS
if (NOS != 1) stop("can only handle 1 response")
is.a.mixed <- if (tmp3.TF[ 2])
rowSums(extra$skip.mix.a) > 0 else rep(FALSE, n)
is.i.mixed <- if (tmp3.TF[ 4])
rowSums(extra$skip.mix.i) > 0 else rep(FALSE, n)
is.d.mixed <- if (tmp3.TF[ 6])
rowSums(extra$skip.mix.d) > 0 else rep(FALSE, n)
is.a.mlmed <- if (tmp3.TF[ 8])
rowSums(extra$skip.mlm.a) > 0 else rep(FALSE, n)
is.i.mlmed <- if (tmp3.TF[ 9])
rowSums(extra$skip.mlm.i) > 0 else rep(FALSE, n)
is.d.mlmed <- if (tmp3.TF[10])
rowSums(extra$skip.mlm.d) > 0 else rep(FALSE, n)
is.ns <- !is.a.mlmed & !is.i.mlmed & !is.d.mlmed &
!is.a.mixed & !is.i.mixed & !is.d.mixed # & !is.truncd
prob.mlm.a <- if (la.mlm) rowSums(pobs.mlm) else 0 # scalar okay
prob.mlm.i <- if (li.mlm) rowSums(pstr.mlm) else 0 # scalar okay
prob.mlm.d <- if (ld.mlm) rowSums(pdip.mlm) else 0 # scalar okay
pmf.deriv1 <- function(y, shape) {
-dzeta(y, shape) * (log(y) +
zeta(shape + 1, deriv = 1) / zeta(shape + 1))
}
pmf.deriv2 <- function(y, shape) {
tmp2 <- zeta(shape + 1, deriv = 1) / zeta(shape + 1)
dzeta(y, shape) * ((log(y) + tmp2)^2 -
zeta(shape + 1, deriv = 2) / zeta(shape + 1) + tmp2^2)
}
sumD.mix.1a.p <- sumD.mix.2a.p <- matrix(0, n, NOS)
if (la.mix > 0) { # \calA_p
DA.mix.0mat.a <- # Matches naming convention further below
DA.mix.1mat.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
sumD.mix.1a.p <- sumD.mix.1a.p + pmf.deriv1(aval, shape.p)
sumD.mix.2a.p <- sumD.mix.2a.p + pmf.deriv2(aval, shape.p)
pmf.a <- dzeta(aval, shape.a)
DA.mix.0mat.a[, jay] <- pmf.a
DA.mix.1mat.a[, jay] <- pmf.deriv1(aval, shape.a)
}
Denom1.a <- rowSums(DA.mix.1mat.a) # aka sumD.mix.1a.a
} # la.mix > 0
if (li.mix) {
DI.mix.0mat.i <- # wrt inflated distribution
DI.mix.1mat.i <- DI.mix.2mat.i <- matrix(0, n, li.mix)
DP.mix.0mat.i <- # wrt parent distribution
DP.mix.1mat.i <- DP.mix.2mat.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.i <- dzeta(ival, shape.i)
DI.mix.0mat.i[, jay] <- pmf.i
DI.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.i)
DI.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.i)
pmf.p <- dzeta(ival, shape.p)
DP.mix.0mat.i[, jay] <- pmf.p
DP.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.p)
DP.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
Denom1.i <- rowSums(DI.mix.1mat.i)
Denom2.i <- rowSums(DI.mix.2mat.i)
} # li.mix
if (ld.mix) {
DD.mix.0mat.d <- # wrt deflated distribution
DD.mix.1mat.d <- DD.mix.2mat.d <- matrix(0, n, ld.mix)
DP.mix.0mat.d <- # wrt parent distribution
DP.mix.1mat.d <- DP.mix.2mat.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.d <- dzeta(dval, shape.d)
DD.mix.0mat.d[, jay] <- pmf.d
DD.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.d)
DD.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.d)
pmf.p <- dzeta(dval, shape.p)
DP.mix.0mat.d[, jay] <- pmf.p
DP.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.p)
DP.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
Denom1.d <- rowSums(DD.mix.1mat.d)
Denom2.d <- rowSums(DD.mix.2mat.d)
} # ld.mix
Bits <- moments.gaitdcombo.zeta(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
sumD.mlm.1a.p <- sumD.mlm.2a.p <- matrix(0, n, NOS)
if (la.mlm)
for (aval in a.mlm) {
sumD.mlm.1a.p <- sumD.mlm.1a.p + pmf.deriv1(aval, shape.p)
sumD.mlm.2a.p <- sumD.mlm.2a.p + pmf.deriv2(aval, shape.p)
}
Denom0.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
Numer <- 1 - pobs.mix - pstr.mix - prob.mlm.a - prob.mlm.i +
pdip.mix + prob.mlm.d
Denom0.a <- c(Bits[["SumA0.mix.a"]]) # Not .p
Denom0.i <- c(Bits[["SumI0.mix.i"]])
Denom0.d <- c(Bits[["SumD0.mix.d"]])
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, NOS)
if (li.mlm > 0) {
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
pmf.p <- dzeta(ival, shape.p)
Dp.mlm.0Mat.i[, jay] <- pmf.p
Dp.mlm.1Mat.i[, jay] <- pmf.deriv1(ival, shape.p)
Dp.mlm.2Mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
} # li.mlm
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, NOS)
if (ld.mlm > 0) {
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
pmf.p <- dzeta(dval, shape.p)
Dp.mlm.0Mat.d[, jay] <- pmf.p
Dp.mlm.1Mat.d[, jay] <- pmf.deriv1(dval, shape.p)
Dp.mlm.2Mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
} # ld.mlm
sumD.1t.p <- sumD.2t.p <-
sumD.1t.a <- sumD.2t.a <-
sumD.1t.i <- sumD.2t.i <-
sumD.1t.d <- sumD.2t.d <- matrix(0, n, NOS)
if (ltruncat)
for (tval in truncate) {
sumD.1t.p <- sumD.1t.p + pmf.deriv1(tval, shape.p)
sumD.2t.p <- sumD.2t.p + pmf.deriv2(tval, shape.p)
sumD.1t.a <- sumD.1t.a + pmf.deriv1(tval, shape.a)
sumD.2t.a <- sumD.2t.a + pmf.deriv2(tval, shape.a)
sumD.1t.i <- sumD.1t.i + pmf.deriv1(tval, shape.i)
sumD.2t.i <- sumD.2t.i + pmf.deriv2(tval, shape.i)
sumD.1t.d <- sumD.1t.d + pmf.deriv1(tval, shape.d)
sumD.2t.d <- sumD.2t.d + pmf.deriv2(tval, shape.d)
}
if (is.finite(max.support)) {
stop("argument 'max.support' must be 'Inf'.")
} # is.finite(max.support)
Denom1.p <- c(-sumD.1t.p - sumD.mlm.1a.p - sumD.mix.1a.p)
Denom2.p <- c(-sumD.2t.p - sumD.mlm.2a.p - sumD.mix.2a.p)
d0B.PI.mlm <- Dp.mlm.0Mat.i / Denom0.p
d1B.PI.mlm <- Dp.mlm.1Mat.i / Denom0.p - # This is most general
Dp.mlm.0Mat.i * Denom1.p / Denom0.p^2
d2B.PI.mlm <- Dp.mlm.2Mat.i / Denom0.p -
2 * Dp.mlm.1Mat.i * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.i * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.i * (Denom1.p^2) / Denom0.p^3
d0B.PD.mlm <- Dp.mlm.0Mat.d / Denom0.p
d1B.PD.mlm <- Dp.mlm.1Mat.d / Denom0.p - # This is most general
Dp.mlm.0Mat.d * Denom1.p / Denom0.p^2
d2B.PD.mlm <- Dp.mlm.2Mat.d / Denom0.p -
2 * Dp.mlm.1Mat.d * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.d * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.d * (Denom1.p^2) / Denom0.p^3
DELTA.i.mlm <- if (li.mlm > 0) {
Numer * d0B.PI.mlm + pstr.mlm # n x li.mlm.
} else {
matrix(0, n, 1) # If li.mlm == 0, for rowSums().
}
DELTA.d.mlm <- if (ld.mlm > 0) {
Numer * d0B.PD.mlm - pdip.mlm # n x ld.mlm.
} else {
matrix(0, n, 1) # If ld.mlm == 0, for rowSums().
}
if (li.mix > 0) {
d0A.i <- DI.mix.0mat.i / Denom0.i
d0B.PI.mix <- DP.mix.0mat.i / Denom0.p
DELTA.i.mix <- Numer * d0B.PI.mix + pstr.mix * d0A.i
d1A.i <- (DI.mix.1mat.i - DI.mix.0mat.i *
Denom1.i / Denom0.i) / Denom0.i
d2A.i <- (DI.mix.2mat.i - (2 * DI.mix.1mat.i * Denom1.i +
DI.mix.0mat.i * Denom2.i) / Denom0.i +
2 * DI.mix.0mat.i * (Denom1.i / Denom0.i)^2) / Denom0.i
d1B.PI.mix <- DP.mix.1mat.i / Denom0.p -
DP.mix.0mat.i * Denom1.p / Denom0.p^2
d2B.PI.mix <- DP.mix.2mat.i / Denom0.p -
2 * DP.mix.1mat.i * Denom1.p / Denom0.p^2 -
DP.mix.0mat.i * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.i * (Denom1.p^2) / Denom0.p^3
} # li.mix > 0
if (ld.mix > 0) {
d0A.d <- DD.mix.0mat.d / Denom0.d
d0B.PD.mix <- DP.mix.0mat.d / Denom0.p
DELTA.d.mix <- Numer * d0B.PD.mix - pdip.mix * d0A.d
d1A.d <- (DD.mix.1mat.d - DD.mix.0mat.d *
Denom1.d / Denom0.d) / Denom0.d
d2A.d <- (DD.mix.2mat.d - (2 * DD.mix.1mat.d * Denom1.d +
DD.mix.0mat.d * Denom2.d) / Denom0.d +
2 * DD.mix.0mat.d * (Denom1.d / Denom0.d)^2) / Denom0.d
d1B.PD.mix <- DP.mix.1mat.d / Denom0.p -
DP.mix.0mat.d * Denom1.p / Denom0.p^2
d2B.PD.mix <- DP.mix.2mat.d / Denom0.p -
2 * DP.mix.1mat.d * Denom1.p / Denom0.p^2 -
DP.mix.0mat.d * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.d * (Denom1.p^2) / Denom0.p^3
} # ld.mix > 0
if (la.mix) {
d0A.a <- DA.mix.0mat.a / Denom0.a
d1A.a <- DA.mix.1mat.a / Denom0.a -
DA.mix.0mat.a * Denom1.a / Denom0.a^2
} # la.mix
fred0.p <- zeta(shape.p + 1)
fred1.p <- zeta(shape.p + 1, deriv = 1)
dl.dshape.p <- -log(y) - fred1.p / fred0.p # Usual formula
dl.dshape.p[!is.ns] <- 0 # For is.a.mixed & is.a.mlmed
dl.dshape.a <- dl.dshape.i <- dl.dshape.d <- numeric(n)
dl.dpstr.mix <- (-1) / Numer # \notin A, I, T, D
dl.dpstr.mix[is.a.mixed] <- 0
dl.dpstr.mix[is.a.mlmed] <- 0
dl.dpdip.mix <- (+1) / Numer # \notin A, I, T, D
dl.dpdip.mix[is.a.mixed] <- 0
dl.dpdip.mix[is.a.mlmed] <- 0
dl.dpobs.mix <- numeric(n) # 0 for \calA_{np}
dl.dpobs.mix[is.ns] <- (-1) / Numer[is.ns]
dl.dpobs.mlm <-
dl.dpstr.mlm <- matrix(0, n, 1) # May be unneeded
dl.dpdip.mlm <- matrix(0, n, max(1, ld.mlm)) # Initzed if used.
dl.dpdip.mlm[is.ns, ] <- 1 / Numer[is.ns]
if (tmp3.TF[ 8] && la.mlm) { # aka \calA_{np}
dl.dpobs.mlm <- matrix(-1 / Numer, n, la.mlm) # \notin calS
dl.dpobs.mlm[!is.ns, ] <- 0 # For a.mix only really
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
is.alt.j.mlm <- extra$skip.mlm.a[, jay] # Logical vector
tmp7a <- 1 / pobs.mlm[is.alt.j.mlm, jay]
dl.dpobs.mlm[is.alt.j.mlm, jay] <- tmp7a
} # jay
} # la.mlm
dl.dshape.p[is.ns] <- dl.dshape.p[is.ns] -
(Denom1.p / Denom0.p)[is.ns]
if (tmp3.TF[ 9] && li.mlm > 0) { # aka \calI_{np}
dl.dpstr.mlm <- matrix(-1 / Numer, n, li.mlm)
dl.dpstr.mlm[!is.ns, ] <- 0 # For a.mlm and a.mix
for (jay in seq(li.mlm)) {
is.inf.j.mlm <- extra$skip.mlm.i[, jay] # Logical vector
tmp7i <- Numer * d1B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
dl.dshape.p[is.inf.j.mlm] <- tmp7i[is.inf.j.mlm]
tmp9i <- d0B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
n.tmp <- -tmp9i[is.inf.j.mlm]
p.tmp <- +tmp9i[is.inf.j.mlm]
if (tmp3.TF[ 8] && la.mlm)
dl.dpobs.mlm[is.inf.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix)
dl.dpobs.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[ 4] && li.mix)
dl.dpstr.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[10] && ld.mlm)
dl.dpdip.mlm[is.inf.j.mlm, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix)
dl.dpdip.mix[is.inf.j.mlm ] <- p.tmp
tmp8 <- (1 - d0B.PI.mlm[, jay]) / DELTA.i.mlm[, jay]
dl.dpstr.mlm[is.inf.j.mlm, ] <- n.tmp # tmp9[is.inf.j.mlm]
dl.dpstr.mlm[is.inf.j.mlm, jay] <- tmp8[is.inf.j.mlm]
} # jay
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # aka \calD_{np}
for (jay in seq(ld.mlm)) {
is.def.j.mlm <- extra$skip.mlm.d[, jay] # Logical vector
tmp7d <- Numer * d1B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
dl.dshape.p[is.def.j.mlm] <- tmp7d[is.def.j.mlm] # 20211020
tmp9d <- d0B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
p.tmp <- +tmp9d[is.def.j.mlm]
n.tmp <- -tmp9d[is.def.j.mlm]
if (tmp3.TF[ 9] && li.mlm)
dl.dpstr.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix)
dl.dpstr.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm)
dl.dpobs.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix)
dl.dpobs.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 6] && ld.mix)
dl.dpdip.mix[is.def.j.mlm ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, jay] <-
dl.dpdip.mlm[is.def.j.mlm, jay] -
1 / DELTA.d.mlm[is.def.j.mlm, jay]
} # jay
} # ld.mlm > 0
if (tmp3.TF[ 2] && la.mix) { # aka \calA_{p}
dl.dpobs.mix[is.a.mixed] <- 1 / pobs.mix[is.a.mixed]
if (tmp3.TF[ 3] && la.mix > 1)
for (jay in seq(la.mix)) {
is.alt.j.mix <- extra$skip.mix.a[, jay] # Logical vector
tmp2 <- d1A.a[, jay] / d0A.a[, jay]
dl.dshape.a[is.alt.j.mix] <- tmp2[is.alt.j.mix] # ccc.
} # jay
} # la.mix
if (tmp3.TF[ 4] && li.mix > 0) { # aka \calI_{p}
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
is.inf.j.mix <- extra$skip.mix.i[, jay] # Logical vector
tmp7b <- Numer * d1B.PI.mix[, jay] / DELTA.i.mix[, jay]
dl.dshape.p[is.inf.j.mix] <- tmp7b[is.inf.j.mix]
tmp8 <- (d0A.i[, jay] - d0B.PI.mix[, jay]) / DELTA.i.mix[, jay]
dl.dpstr.mix[is.inf.j.mix] <- tmp8[is.inf.j.mix]
if (li.mix > 1) {
tmp2 <- pstr.mix * d1A.i[, jay] / DELTA.i.mix[, jay]
dl.dshape.i[is.inf.j.mix] <- tmp2[is.inf.j.mix]
}
tmp9i <- d0B.PI.mix[, jay] / DELTA.i.mix[, jay]
n.tmp <- -tmp9i[is.inf.j.mix]
p.tmp <- +tmp9i[is.inf.j.mix]
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mix, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mix ] <- p.tmp
} # jay
} # li.mix > 0
if (tmp3.TF[ 6] && ld.mix > 0) { # aka \calD_{p}
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
is.def.j.mix <- extra$skip.mix.d[, jay] # Logical vector
tmp7b <- Numer * d1B.PD.mix[, jay] / DELTA.d.mix[, jay]
dl.dshape.p[is.def.j.mix] <- tmp7b[is.def.j.mix]
tmp8 <- (d0B.PD.mix[, jay] - d0A.d[, jay]) / DELTA.d.mix[, jay]
dl.dpdip.mix[is.def.j.mix] <- tmp8[is.def.j.mix]
if (ld.mix > 1) {
tmp2 <- (-pdip.mix) * d1A.d[, jay] / DELTA.d.mix[, jay]
dl.dshape.d[is.def.j.mix] <- tmp2[is.def.j.mix]
}
tmp9d <- d0B.PD.mix[, jay] / DELTA.d.mix[, jay]
n.tmp <- -tmp9d[is.def.j.mix]
p.tmp <- +tmp9d[is.def.j.mix]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.def.j.mix, ] <- p.tmp
} # jay
} # ld.mix > 0
new.ansd <- matrix(0, n, M) # Same dimension as eta
tmp3.TF <- !is.na(rowSums(extra$indeta))
if (lall.len) { # An MLM fitted
all6.dldp <- cbind(if (tmp3.TF[ 2]) dl.dpobs.mix else NULL,
if (tmp3.TF[ 4]) dl.dpstr.mix else NULL,
if (tmp3.TF[ 6]) dl.dpdip.mix else NULL,
if (tmp3.TF[ 8]) dl.dpobs.mlm else NULL,
if (tmp3.TF[ 9]) dl.dpstr.mlm else NULL,
if (tmp3.TF[10]) dl.dpdip.mlm else NULL)
rSs.tmp <- rowSums(allprobs[, -ncol(allprobs), drop = FALSE] *
all6.dldp)
new.ansd[, -ind.shape.z] <- allprobs[, -ncol(allprobs)] *
(all6.dldp - rSs.tmp)
} # lall.len
dshape.p.deta <- dtheta.deta(shape.p, .lshape.p , .eshape.p )
if (tmp3.TF[ 3])
dshape.a.deta <- dtheta.deta(shape.a, .lshape.a , .eshape.a )
if (tmp3.TF[ 5])
dshape.i.deta <- dtheta.deta(shape.i, .lshape.i , .eshape.i )
if (tmp3.TF[ 7])
dshape.d.deta <- dtheta.deta(shape.d, .lshape.d , .eshape.d )
new.ansd[, 1] <- dl.dshape.p * dshape.p.deta
if (tmp3.TF[ 3])
new.ansd[, extra$indeta[3, 1]] <- dl.dshape.a * dshape.a.deta
if (tmp3.TF[ 5])
new.ansd[, extra$indeta[5, 1]] <- dl.dshape.i * dshape.i.deta
if (tmp3.TF[ 7])
new.ansd[, extra$indeta[7, 1]] <- dl.dshape.d * dshape.d.deta
onecoln.indeta <- extra$indeta[1:7, ] # One coln params only
onecoln.indeta <- na.omit(onecoln.indeta) # Only those present
allcnames <- c(rownames(onecoln.indeta),
as.character(c(a.mlm, i.mlm, d.mlm)))
colnames(new.ansd) <- allcnames
c(w) * new.ansd
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.truncate = truncate, .max.support = max.support ))),
weight = eval(substitute(expression({ # gaitdzeta
wz <- matrix(0, n, M * (M + 1) / 2) # The complete size
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom0.p)
if (min(probns) < 0 || 1 < max(probns))
stop("variable 'probns' for P(nonspecial) is out of range")
zero0n <- numeric(n)
ned2l.dpobs.mix.shape.p <- zero0n # mB overwritten below [4279]
ned2l.dpobs.mix.shape.a <- zero0n # Fini; (2, 3) element
ned2l.dpobs.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpobs.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.p <- zero0n # Optional (1, 4) element
ned2l.dpstr.mix.shape.a <- zero0n # Final; nothing to do
ned2l.dpstr.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpdip.mix.shape.p <- zero0n # Optional (1, 6) element
posn.pobs.mix <- as.vector(extra$indeta[ 2, 'launch'])
posn.shape.a <- as.vector(extra$indeta[ 3, 'launch'])
posn.pstr.mix <- as.vector(extra$indeta[ 4, 'launch'])
posn.shape.i <- as.vector(extra$indeta[ 5, 'launch'])
posn.pdip.mix <- as.vector(extra$indeta[ 6, 'launch'])
posn.shape.d <- as.vector(extra$indeta[ 7, 'launch'])
posn.pobs.mlm <- as.vector(extra$indeta[ 8, 'launch'])
posn.pstr.mlm <- as.vector(extra$indeta[ 9, 'launch'])
posn.pdip.mlm <- as.vector(extra$indeta[10, 'launch'])
ned2l.dpdip.mix2 <- # Elt (6, 6)
ned2l.dpstr.mix2 <- # Elt (4, 4). Unchanged by deflation.
ned2l.dpobs.mlm.pstr.mix <- # Elts (4, >=8). (((09)))
ned2l.dpobs.mix.pstr.mix <- +probns / Numer^2 # ccc Elt (2, 4)
if (all(c(la.mix, li.mlm) > 0)) # (((08)))
ned2l.dpobs.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(li.mix, li.mlm) > 0)) # (((10)))
ned2l.dpstr.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(ld.mix, ld.mlm) > 0)) # (((21)))
ned2l.dpdip.mix.pdip.mlm <- matrix( probns / Numer^2, n, ld.mlm)
ned2l.dpobs.mlm.pdip.mix <- # Elts (6, >=8). (((19)))
ned2l.dpstr.mix.pdip.mix <- # Elt (4, 6)
ned2l.dpobs.mix.pdip.mix <- -probns / Numer^2 # ccc Elt (2, 6)
if (all(c(la.mix, ld.mlm) > 0)) # (((17)))
ned2l.dpobs.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(li.mix, ld.mlm) > 0)) # (((18)))
ned2l.dpstr.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(ld.mix, li.mlm) > 0)) # (((20)))
ned2l.dpdip.mix.pstr.mlm <- matrix(-probns / Numer^2, n, li.mlm)
ned2l.dshape.p2 <- probns * (
zeta(shape.p + 1, deriv = 2) / fred0.p -
(fred1.p / fred0.p)^2 + # ccc
Denom2.p / Denom0.p - (Denom1.p / Denom0.p)^2) +
(if (tmp3.TF[ 4] && li.mix) Numer *
rowSums(Numer * (d1B.PI.mix^2) / DELTA.i.mix - d2B.PI.mix) else 0) +
(if (tmp3.TF[ 9] && li.mlm) Numer *
rowSums(Numer * (d1B.PI.mlm^2) / DELTA.i.mlm - d2B.PI.mlm) else 0) +
(if (tmp3.TF[ 6] && ld.mix) Numer *
rowSums(Numer * (d1B.PD.mix^2) / DELTA.d.mix - d2B.PD.mix) else 0) +
(if (tmp3.TF[10] && ld.mlm) Numer * # nnn.
rowSums(Numer * (d1B.PD.mlm^2) / DELTA.d.mlm - d2B.PD.mlm) else 0)
wz[, iam(1, 1, M)] <- ned2l.dshape.p2 * dshape.p.deta^2
ned2l.dpobs.mix2 <- 1 / pobs.mix + probns / Numer^2
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpobs.mix2 <- # More just below, ccc
ned2l.dpobs.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpobs.mix2 <- # ccc.
ned2l.dpobs.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (tmp3.TF[ 6] && ld.mix > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (tmp3.TF[ 2] && la.mix > 0)
wz[, iam(2, 2, M)] <- ned2l.dpobs.mix2 # Link done later
if (tmp3.TF[ 3] && la.mix > 1) {
ned2l.dshape.a2 <- pobs.mix * (
rowSums((DA.mix.1mat.a^2) / DA.mix.0mat.a) / Denom0.a -
(Denom1.a / Denom0.a)^2) # ccc.
wz[, iam(3, 3, M)] <- ned2l.dshape.a2 * dshape.a.deta^2
}
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 +
rowSums((d0A.i - d0B.PI.mix)^2 / DELTA.i.mix)
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mix * (1 + Numer * (d0A.i - d0B.PI.mix) / DELTA.i.mix))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (all(tmp3.TF[c(2, 4)]))
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix +
rowSums(-d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
} # (tmp3.TF[ 4] && li.mix > 0)
if (all(tmp3.TF[c(2, 4, 9)])) { # was la.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 4, 6)])) { # == ld.mix > 0 & DELTA.d.mix
ned2l.dpobs.mix.pstr.mix <- # nnn
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(2, 4, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pstr.mix <- # nnn.
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pstr.mix))
wz[, iam(posn.pobs.mix, posn.pstr.mix, M)] <-
ned2l.dpobs.mix.pstr.mix # Link done later
if (all(tmp3.TF[c(2, 6)]))
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix +
rowSums( d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
if (all(tmp3.TF[c(2, 6, 9)])) { # == li.mlm > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 6, 4)])) { # == li.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (all(tmp3.TF[c(2, 6, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pdip.mix <- # nnn.
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pdip.mix))
wz[, iam(posn.pobs.mix, posn.pdip.mix, M)] <-
ned2l.dpobs.mix.pdip.mix # Link done later
if (tmp3.TF[ 5] && li.mix > 1) { # \calI_{p}, includes \theta_i.
ned2l.dshape.p.shape.i <- pstr.mix * Numer *
rowSums(d1A.i * d1B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(1, posn.shape.i, M)] <- ned2l.dshape.p.shape.i *
dshape.p.deta * dshape.i.deta # All links done here
ned2l.dshape.i2 <- pstr.mix *
rowSums(pstr.mix * (d1A.i^2) / DELTA.i.mix - d2A.i) # ccc.
wz[, iam(posn.shape.i, posn.shape.i, M)] <-
ned2l.dshape.i2 * dshape.i.deta^2
if (tmp3.TF[ 2]) { # tmp3.TF[ 4] is TRUE, given tmp3.TF[ 5]
ned2l.dpobs.mix.shape.i <-
rowSums(-pstr.mix * d1A.i * d0B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(posn.pobs.mix, posn.shape.i, M)] <-
ned2l.dpobs.mix.shape.i # * dshape.i.deta done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.i <- rowSums( # ccc.
d1A.i * (pstr.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix - 1))
wz[, iam(posn.pstr.mix, posn.shape.i, M)] <-
ned2l.dpstr.mix.shape.i # * dshape.i.deta done later
}
if (all(tmp3.TF[c(5, 6)])) {
ned2l.dpdip.mix.shape.i <- rowSums(
(-pstr.mix) * d0B.PI.mix * d1A.i / DELTA.i.mix)
wz[, iam(posn.pdip.mix, posn.shape.i, M)] <-
ned2l.dpdip.mix.shape.i # link done later
}
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.i <- rowSums(
-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix) # ccc.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.i, M)] <-
ned2l.dpobs.mlm.shape.i # * dshape.i.deta done later
}
} # (tmp3.TF[ 5] && li.mix > 1)
if (tmp3.TF[ 6] && ld.mix > 0) { # \calD_{p}, maybe w. \theta_d
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mix * (1 + Numer * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 +
rowSums((d0A.d - d0B.PD.mix)^2 / DELTA.d.mix)
} # (tmp3.TF[ 6] && ld.mix > 0)
if (tmp3.TF[ 7] && ld.mix > 1) { # \calD_{p}, includes \theta_d
ned2l.dshape.p.shape.d <- (-pdip.mix) * Numer *
rowSums(d1A.d * d1B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(1, posn.shape.d, M)] <- ned2l.dshape.p.shape.d *
dshape.p.deta * dshape.d.deta # All links done here
if (tmp3.TF[ 2]) { # tmp3.TF[ 6] is TRUE, given tmp3.TF[ 7]
ned2l.dpobs.mix.shape.d <-
rowSums(pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(posn.pobs.mix, posn.shape.d, M)] <-
ned2l.dpobs.mix.shape.d # link done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.d <- rowSums(
pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix)
wz[, iam(posn.pstr.mix, posn.shape.d, M)] <-
ned2l.dpstr.mix.shape.d # * dshape.i.deta done later
}
ned2l.dpdip.mix.shape.d <- rowSums(
d1A.d * (1 + pdip.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
wz[, iam(posn.pdip.mix, posn.shape.d, M)] <-
ned2l.dpdip.mix.shape.d # * dshape.d.deta done later
ned2l.dshape.d2 <- pdip.mix *
rowSums(pdip.mix * (d1A.d^2) / DELTA.d.mix + d2A.d) # nnn.
wz[, iam(posn.shape.d, posn.shape.d, M)] <-
ned2l.dshape.d2 * dshape.d.deta^2
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.d <- rowSums(
pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix) # nnn.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.d, M)] <-
ned2l.dpobs.mlm.shape.d # * dshape.d.deta done later
}
} # (tmp3.TF[ 7] && ld.mix > 1)
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # ccc
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ned2l.dpstr.mix.shape.p <- # ccc.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
} # tmp3.TF[ 9] && li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # nnn.
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpstr.mix.shape.p <- # nnn.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
} # tmp3.TF[10] && ld.mlm > 0
if (!is.na(posn.pobs.mix)) # Optional (1, 2) element:
wz[, iam(1, posn.pobs.mix, M)] <-
ned2l.dpobs.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix)) # Optional (1, 4) element
wz[, iam(1, posn.pstr.mix, M)] <-
ned2l.dpstr.mix.shape.p # One link done later
if (!is.na(posn.pdip.mix)) # Optional (1, 6) element
wz[, iam(1, posn.pdip.mix, M)] <-
ned2l.dpdip.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix) &&
!is.na(posn.pdip.mix)) # Optional (4, 6) element
wz[, iam(posn.pstr.mix, posn.pdip.mix, M)] <-
ned2l.dpstr.mix.pdip.mix # Links done later zz1
if (!is.na(posn.pstr.mix)) # Optional (4, 4) element
wz[, iam(posn.pstr.mix, # Link done later
posn.pstr.mix, M)] <- ned2l.dpstr.mix2
if (!is.na(posn.pdip.mix)) # Optional (6, 6) element
wz[, iam(posn.pdip.mix, # Link done later
posn.pdip.mix, M)] <- ned2l.dpdip.mix2
if (tmp3.TF[ 8] && la.mlm) { # \calA_{np}, includes \omega_s
ofset <- posn.pobs.mlm - 1 # 7 for GAITD combo
for (uuu in seq(la.mlm)) { # Diagonal elts only
wz[, iam(ofset + uuu,
ofset + uuu, M)] <- 1 / pobs.mlm[, uuu]
} # uuu
tmp8a <- probns / Numer^2
if (tmp3.TF[ 4] && li.mix)
tmp8a <- tmp8a + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
tmp8a <- tmp8a + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
tmp8a <- tmp8a + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
tmp8a <- tmp8a + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # All elts
for (vvv in uuu:la.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
wz[, iam(ofset + uuu, ofset + vvv, M)] + tmp8a # All elts
} # la.mlm
if (tmp3.TF[ 8] && la.mlm) {
init0.i.val <- init0.d.val <- 0
if (tmp3.TF[ 9] && li.mlm) init0.i.val <-
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[10] && ld.mlm) init0.d.val <-
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpobs.mlm.shape.p <- init0.i.val + init0.d.val # Vector
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums( # nnn
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ofset <- posn.pobs.mlm - 1 # 5 for combo
for (vvv in seq(la.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpobs.mlm.shape.p
} # la.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s
init0.val <- probns / Numer^2
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (ld.mix) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (ld.mlm) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
ned2l.dpstr.mlm2 <-
matrix(init0.val, n, li.mlm * (li.mlm + 1) / 2)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss])^2 / DELTA.i.mlm[, sss]
if (li.mlm > 1) {
for (uuu in seq(li.mlm - 1))
for (vvv in (uuu + 1):li.mlm)
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
} # if (li.mlm > 1)
ofset <- posn.pstr.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in uuu:li.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s
init0.val <- probns / Numer^2
if (ld.mix)
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (li.mlm)
init0.val <- init0.val + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
ned2l.dpdip.mlm2 <-
matrix(init0.val, n, ld.mlm * (ld.mlm + 1) / 2)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu))^2 / DELTA.d.mlm[, sss]
if (ld.mlm > 1) {
for (uuu in seq(ld.mlm - 1))
for (vvv in (uuu + 1):ld.mlm)
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu)) *
(d0B.PD.mlm[, sss] - (sss == vvv)) / DELTA.d.mlm[, sss]
} # if (ld.mlm > 1)
ofset <- posn.pdip.mlm - 1
for (uuu in seq(ld.mlm))
for (vvv in uuu:ld.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)]
} # ld.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpstr.mlm.theta.p <- matrix(0, n, li.mlm)
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.theta.p[, vvv] <-
ned2l.dpstr.mlm.theta.p[, vvv] +
d1B.PI.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PI.mlm[, sss]) / (
DELTA.i.mlm[, sss]))
if (li.mix && tmp3.TF[ 4])
ned2l.dpstr.mlm.theta.p <-
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpstr.mlm.theta.p <- # nnn
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (ld.mlm && tmp3.TF[10])
ned2l.dpstr.mlm.theta.p <- # nnn.
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ofset <- posn.pstr.mlm - 1
for (vvv in seq(li.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpstr.mlm.theta.p[, vvv]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpdip.mlm.theta.p <- matrix(0, n, ld.mlm)
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm.theta.p[, vvv] <-
ned2l.dpdip.mlm.theta.p[, vvv] - # Minus
d1B.PD.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PD.mlm[, sss]) / (
DELTA.d.mlm[, sss]))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (li.mix && tmp3.TF[ 4])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (li.mlm && tmp3.TF[ 9])
ned2l.dpdip.mlm.theta.p <- # nnn.
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ofset <- posn.pdip.mlm - 1
for (vvv in seq(ld.mlm)) # nnn.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpdip.mlm.theta.p[, vvv]
} # ld.mlm > 0
if (li.mlm && li.mix > 1) {
ned2l.dpstr.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.i, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.i # ccc.
} # li.mlm && li.mix > 1
if (ld.mlm && ld.mix > 1) {
ned2l.dpdip.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.d, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.d # nnn.
} # ld.mlm && ld.mix > 1
if (ld.mlm && li.mix > 1) {
ned2l.dpdip.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.i, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.i # nnn.
} # ld.mlm && li.mix > 1
if (li.mlm && ld.mix > 1) {
ned2l.dpstr.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.d, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.d # nnn.
} # li.mlm && ld.mix > 1
if (all(c(la.mlm, li.mlm) > 0)) {
ned2l.dpobs.mlm.pstr.mlm <-
array(probns / Numer^2, c(n, la.mlm, li.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] - d0B.PI.mlm[, sss] *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pstr.mlm <-
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pstr <- posn.pstr.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pstr + vvv, M)] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(li.mlm, ld.mlm) > 0)) {
ned2l.dpstr.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, li.mlm, ld.mlm))
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PI.mlm[, sss] *
((sss == uuu) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpstr.mlm.pdip.mlm <-
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpstr.mlm.pdip.mlm <- # nnn.
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pstr <- posn.pstr.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pstr + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv]
} # all(c(li.mlm, ld.mlm) > 0)
if (all(c(la.mlm, ld.mlm) > 0)) {
ned2l.dpobs.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, la.mlm, ld.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(la.mix, la.mlm) > 0)) {
ned2l.dpobs.mix.pobs.mlm <- probns / Numer^2 # Initialize
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 7]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pobs.mix, posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pobs.mlm # Link done later
}
if (all(c(la.mix, li.mlm) > 0)) { # all(tmp3.TF[c(2, 9)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pstr.mlm <-
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pstr.mlm <- # nnn
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mix.pstr.mlm <- # nnn; + is correct, not -
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mix.pstr.mlm[, uuu] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm)) # ccc.
wz[, iam(posn.pobs.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] # Link done later
} # all(c(la.mix, li.mlm) > 0)
if (all(c(la.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(2, 10)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mix.pdip.mlm[, uuu] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
for (uuu in seq(ld.mlm)) # nnn.
wz[, iam(posn.pobs.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] # Link done later
} # all(c(la.mix, ld.mlm) > 0)
if (all(c(li.mix, la.mlm) > 0)) { # all(tmp3.TF[c(4, 8)])
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pstr.mix <-
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ned2l.dpobs.mlm.pstr.mix <- # tmp3.TF[ 4] && li.mix
ned2l.dpobs.mlm.pstr.mix -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pstr.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pstr.mix # Link done later
} # all(c(li.mix, la.mlm) > 0
if (all(c(ld.mix, la.mlm) > 0)) { # all(tmp3.TF[c(6, 8)])
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
ned2l.dpobs.mlm.pdip.mix <- # all(tmp3.TF[c(6, 8)])
ned2l.dpobs.mlm.pdip.mix +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
for (uuu in seq(la.mlm)) # nnn.
wz[, iam(posn.pdip.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pdip.mix # Link done later
} # all(c(ld.mix, la.mlm) > 0
if (all(c(li.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpstr.mix.pstr.mlm[, uuu] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
ned2l.dpstr.mix.pstr.mlm <-
ned2l.dpstr.mix.pstr.mlm -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] # Link done later
} # all(c(li.mix, li.mlm) > 0
if (all(c(ld.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(6, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[ 9]
for (sss in seq(ld.mlm))
ned2l.dpdip.mix.pdip.mlm[, uuu] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] -
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm -
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] # Link done later
} # all(c(ld.mix, ld.mlm) > 0
if (all(c(ld.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpdip.mix.pstr.mlm[, uuu] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] # Link done later
} # all(c(ld.mix, li.mlm) > 0
if (all(c(li.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(4, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[10]
for (sss in seq(ld.mlm))
ned2l.dpstr.mix.pdip.mlm[, uuu] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (li.mix) # tmp3.TF[ 4]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm +
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpstr.mix.pdip.mlm <- # nnn.
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] # Link done later
} # all(c(li.mix, ld.mlm) > 0)
if (lall.len) {
wz.6 <- matrix(0, n, M * (M + 1) / 2) # Or == 0 * wz
ind.rc <- setdiff(1:M, ind.shape.z) # Contiguous rows and
lind.rc <- length(ind.rc) # cols of the DAMLM
# Copy in the thetas values: the looping is overkill.
for (uuu in ind.shape.z)
for (sss in seq(M))
wz.6[, iam(uuu, sss, M)] <- wz[, iam(uuu, sss, M)]
speed.up <- intercept.only && (
length(offset) == 1 || all(offset[1] == offset))
IND.mlm <- iam(NA, NA, lind.rc, both = TRUE, diag = TRUE)
n.use <- if (speed.up) 2 else n # For sandwich.mlm
if (!length(extra$ind.wz.match)) {
Imat <- matrix(NA, lind.rc, lind.rc)
for (jay in seq(lind.rc)) {
iptr <- jay
for (kay in (ind.rc[jay]):M) {
if (!any(kay %in% ind.shape.z)) {
Imat[jay, iptr] <-
which(extra$index.M$row == ind.rc[jay] &
extra$index.M$col == kay)
iptr <- iptr + 1
} # if
} # kay
} # jay
ind.wz.match <- Imat[cbind(IND.mlm$row.ind,
IND.mlm$col.ind)]
extra$ind.wz.match <- ind.wz.match # Assign it once
} # !length(extra$ind.wz.match)
filling <- if (speed.up)
wz[1:n.use, extra$ind.wz.match, drop = FALSE] else
wz[, extra$ind.wz.match, drop = FALSE]
M.mlm <- lind.rc
if (is.null(extra$iamlist)) {
extra$iamlist <- iamlist <-
iam(NA, NA, M = M.mlm, both = TRUE)
if (M.mlm > 1) { # Offdiagonal elts
extra$iamlist.nod <- iamlist.nod <-
iam(NA, NA, M.mlm, both = TRUE, diag = FALSE)
}
} # is.null(extra$iamlist)
iamlist <- extra$iamlist
iamlist.nod <- extra$iamlist.nod
MM12.mlm <- M.mlm * (M.mlm + 1) / 2
Qf3 <- rowSums(filling[, 1:M.mlm, drop = FALSE] * # Diag elts
(allprobs[1:n.use, 1:M.mlm, drop = FALSE])^2)
if (M.mlm > 1) # Offdiagonal elts
Qf3 <- Qf3 + 2 * rowSums(allprobs[1:n.use, iamlist.nod$row] *
filling[, -(1:M.mlm), drop = FALSE] * # n-vector
allprobs[1:n.use, iamlist.nod$col])
Qf3 <- matrix(Qf3, n.use, MM12.mlm)
Qf2rowsums <- matrix(0, n.use, M.mlm) # rowsums stored colwise
for (want in seq(M.mlm)) { # Want the \equiv of rowSums(Qf2a)
iamvec <- iam(want, 1:M.mlm, M = M.mlm) # Diagonals included
Qf2rowsums[, want] <-
rowSums(filling[, iamvec, drop = FALSE] *
allprobs[1:n.use, 1:M.mlm])
} # want
Qf2a <- Qf2rowsums[, iamlist$row]
Qf2b <- Qf2rowsums[, iamlist$col]
Qform <- filling - Qf2a - Qf2b + Qf3 # n x MM12.mlm
Qform <- Qform *
allprobs[1:n.use, iamlist$row, drop = FALSE] *
allprobs[1:n.use, iamlist$col, drop = FALSE]
wz.6[, extra$ind.wz.match] <- if (speed.up)
matrix(Qform[1, ], n, ncol(Qform), byrow = TRUE) else c(Qform)
dstar.deta <- cbind(dshape.p.deta,
if (tmp3.TF[ 3]) dshape.a.deta else NULL,
if (tmp3.TF[ 5]) dshape.i.deta else NULL,
if (tmp3.TF[ 7]) dshape.d.deta else NULL)
iptr <- 0
if (length(ind.shape.z))
for (uuu in ind.shape.z) { # Could delete 3 4 shape.a (orthog)
iptr <- iptr + 1
for (ttt in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <- 0 # Initialize
for (sss in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <-
wz.6[, iam(uuu, ind.rc[ttt], M)] +
allprobs[, sss] * (max(0, sss == ttt) -
allprobs[, ttt]) *
wz[, iam(uuu, ind.rc[sss], M)] * dstar.deta[, iptr]
} # sss
} # ttt
} # uuu
wz <- wz.6 # Completed
} # lall.len
if (lall.len) { # A MLM was fitted
mytiny <- (allprobs < sqrt(.Machine$double.eps)) |
(allprobs > 1.0 - sqrt(.Machine$double.eps))
atiny <- rowSums(mytiny) > 0
if (any(atiny)) {
ind.diags <- setdiff(1:M, ind.shape.z) # Exclude thetas
wz[atiny, ind.diags] <- .Machine$double.eps +
wz[atiny, ind.diags] * (1 + .Machine$double.eps^0.5)
}
} # lall.len
c(w) * wz
}), list( .truncate = truncate ))))
} # gaitdzeta
gaitdlog <-
function(a.mix = NULL, i.mix = NULL,
d.mix = NULL,
a.mlm = NULL, i.mlm = NULL, # Unstructured probs are
d.mlm = NULL, # contiguous
truncate = NULL, max.support = Inf,
zero = c("pobs", "pstr", "pdip"), # Pruned, handles all 6
eq.ap = TRUE, eq.ip = TRUE, eq.dp = TRUE,
parallel.a = FALSE, parallel.i = FALSE,
parallel.d = FALSE,
lshape.p = "logitlink",
lshape.a = lshape.p, # "logitlink", 20201117
lshape.i = lshape.p, # "logitlink", 20201117
lshape.d = lshape.p, # "logitlink", 20211011
type.fitted = c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"),
gshape.p = -expm1(-7 * ppoints(12)),
gpstr.mix = ppoints(7) / 3, # ppoints(9) / 2,
gpstr.mlm = ppoints(7) / (3 + length(i.mlm)),
imethod = 1,
mux.init = c(0.75, 0.5, 0.75), # Order is A, I, D.
ishape.p = NULL, ishape.a = ishape.p,
ishape.i = ishape.p, ishape.d = ishape.p,
ipobs.mix = NULL, ipstr.mix = NULL, # 0.25,
ipdip.mix = NULL, # 0.01, # Easy but inflexible 0.01
ipobs.mlm = NULL, ipstr.mlm = NULL, # 0.25,
ipdip.mlm = NULL, # 0.01, # NULL, Easy but inflexible
byrow.aid = FALSE,
ishrinkage = 0.95,
probs.y = 0.35) {
mux.init <- rep_len(mux.init, 3)
if (length(a.mix) == 0) a.mix <- NULL
if (length(i.mix) == 0) i.mix <- NULL
if (length(d.mix) == 0) d.mix <- NULL
if (length(a.mlm) == 0) a.mlm <- NULL
if (length(i.mlm) == 0) i.mlm <- NULL
if (length(d.mlm) == 0) d.mlm <- NULL
if (length(truncate) == 0) truncate <- NULL
lowsup <- 1
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support,
min.support = lowsup)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltruncat <- length(truncate <- sort(truncate))
ltrunc.use <- ltruncat > 0 || !is.infinite(max.support)
lshape.p <- as.list(substitute(lshape.p))
eshape.p <- link2list(lshape.p)
lshape.p <- attr(eshape.p, "function.name")
lshape.p.save <- lshape.p
lpobs.mix <- "multilogitlink" # \omega_p
epobs.mix <- list() # zz NULL for now 20200907 coz 'multilogitlink'
eshape.a <- link2list(lshape.a)
lshape.a <- attr(eshape.a, "function.name")
lpstr.mix <- "multilogitlink" # \phi_p
epstr.mix <- list() # zz NULL for now 20200907 coz 'multilogitlink'
lpdip.mix <- "multilogitlink" # zz unsure 20211002
epdip.mix <- list() # zz unsure 20211002
eshape.i <- link2list(lshape.i)
lshape.i <- attr(eshape.i, "function.name")
eshape.d <- link2list(lshape.d)
lshape.d <- attr(eshape.d, "function.name")
lshape.p.save <- lshape.p
gshape.p.save <- gshape.p
if (is.vector(zero) && is.character(zero) && length(zero) == 3) {
if (li.mix + li.mlm == 0)
zero <- setdiff(zero, "pstr")
if (la.mix + la.mlm == 0)
zero <- setdiff(zero, "pobs")
if (ld.mix + ld.mlm == 0)
zero <- setdiff(zero, "pdip")
if (length(zero) == 0)
zero <- NULL # Better than character(0)
}
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
if (lall.len + ltruncat == 0 && is.infinite(max.support))
return(eval(substitute(
logff(lshape = .lshape.p.save ,
gshape = .gshape.p.save ,
zero = NULL),
list( .lshape.p.save = lshape.p.save,
.gshape.p.save = gshape.p.save ))))
if (!is.logical(eq.ap) || length(eq.ap) != 1)
stop("argument 'eq.ap' must be a single logical")
if (!is.logical(eq.ip) || length(eq.ip) != 1)
stop("argument 'eq.ip' must be a single logical")
if (!is.logical(parallel.a) || length(parallel.a) != 1)
stop("argument 'parallel.a' must be a single logical")
if (!is.logical(parallel.i) || length(parallel.i) != 1)
stop("argument 'parallel.i' must be a single logical")
if (!is.logical(parallel.d) || length(parallel.d) != 1)
stop("argument 'parallel.d' must be a single logical")
if (FALSE) { # Comment this out to allow default eq.ap = TRUE, etc.
if (la.mix <= 1 && eq.ap)
stop("<= one unstructured altered value (no 'shape.a')",
", so setting 'eq.ap = TRUE' is meaningless")
if (li.mix <= 1 && eq.ip)
stop("<= one unstructured inflated value (no 'shape.i')",
", so setting 'eq.ip = TRUE' is meaningless")
if (ld.mix <= 1 && eq.dp)
stop("<= one unstructured deflated value (no 'shape.d')",
", so setting 'eq.dp = TRUE' is meaningless")
if (la.mlm <= 1 && parallel.a) # Only \omega_1
stop("<= one altered mixture probability, 'pobs", a.mlm,
"', so setting 'parallel.a = TRUE' is meaningless")
if (li.mlm <= 1 && parallel.i) # Only \phi_1
stop("<= one inflated mixture probability, 'pstr", i.mlm,
"', so setting 'parallel.i = TRUE' is meaningless")
if (ld.mlm <= 1 && parallel.d) # Only \psi_1
stop("<= one deflated mixture probability, 'pdip", d.mlm,
"', so setting 'parallel.d = TRUE' is meaningless")
} # FALSE
type.fitted.choices <-
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s")
type.fitted <- match.arg(type.fitted[1], type.fitted.choices)[1]
tmp7a <- if (la.mlm) paste0("pobs.mlm", a.mlm) else NULL
tmp7b <- if (li.mlm) paste0("pstr.mlm", i.mlm) else NULL
tmp7c <- if (ld.mlm) paste0("pdip.mlm", d.mlm) else NULL
tmp3 <- c(shape.p = lshape.p,
pobs.mix = if (la.mix) "multilogitlink" else NULL,
shape.a = if (la.mix > 1) lshape.a else NULL,
pstr.mix = if (li.mix) "multilogitlink" else NULL,
shape.i = if (li.mix > 1) lshape.i else NULL,
pdip.mix = if (ld.mix) "multilogitlink" else NULL,
shape.d = if (ld.mix > 1) lshape.d else NULL,
if (la.mlm) rep("multilogitlink", la.mlm) else NULL,
if (li.mlm) rep("multilogitlink", li.mlm) else NULL,
if (ld.mlm) rep("multilogitlink", ld.mlm) else NULL)
Ltmp3 <- length(tmp3)
if (la.mlm + li.mlm + ld.mlm)
names(tmp3)[(Ltmp3 - la.mlm - li.mlm - ld.mlm + 1):Ltmp3] <-
c(tmp7a, tmp7b, tmp7c)
par1or2 <- 1 # 2
tmp3.TF <- c(TRUE, la.mix > 0, la.mix > 1,
li.mix > 0, li.mix > 1,
ld.mix > 0, ld.mix > 1,
la.mlm > 0, li.mlm > 0, ld.mlm > 0)
indeta.finish <- cumsum(c(par1or2, 1, par1or2,
1, par1or2,
1, par1or2,
la.mlm, li.mlm, ld.mlm,
ld.mlm + 1) * c(tmp3.TF, 1))
indeta.launch <- c(1, 1 + head(indeta.finish, -1))
indeta.launch <- head(indeta.launch, -1)
indeta.finish <- head(indeta.finish, -1)
indeta.launch[!tmp3.TF] <- NA # Not to be accessed
indeta.finish[!tmp3.TF] <- NA # Not to be accessed
indeta <- cbind(launch = indeta.launch,
finish = indeta.finish)
rownames(indeta) <- c("shape.p",
"pobs.mix", "shape.a",
"pstr.mix", "shape.i",
"pdip.mix", "shape.d",
"pobs.mlm", "pstr.mlm", "pdip.mlm")
M1 <- max(indeta, na.rm = TRUE)
predictors.names <- tmp3 # Passed into @infos and @initialize.
blurb1 <- "L" # zz1
if (la.mlm + la.mix) blurb1 <- "Generally-altered l"
if (li.mlm + li.mix) blurb1 <- "Generally-inflated l"
if (ltrunc.use) blurb1 <- "Generally-truncated l"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && !ltrunc.use)
blurb1 <- "Generally-altered and -inflated l"
if ( (la.mlm + la.mix) && !(li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered and -truncated l"
if (!(la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-inflated and -truncated l"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered, -inflated and -truncated l"
if (ld.mlm + ld.mix) blurb1 <-
c(blurb1,
if (la.mlm + la.mix + li.mlm + li.mix) "and " else "Generally",
"-deflated ")
new("vglmff",
blurb = c(blurb1, "ogarithmic regression\n",
"(GAITD-Log(shape.p)-",
"Log(shape.a)-MLM-",
"Log(shape.i)-MLM-\n",
"Log(shape.d)-MLM generally)\n\n",
"Links: ",
namesof("shape.p", lshape.p, earg = eshape.p,
tag = FALSE),
if (la.mix > 0) c(", ", "multilogit(pobs.mix)"),
if (la.mix > 1) c(", ",
namesof("shape.a", lshape.a, eshape.a, tag = FALSE)),
if (la.mix && li.mix) ", \n ",
if (li.mix > 0) c( if (la.mix) "" else ", ",
"multilogit(pstr.mix)"),
if (li.mix > 1) c(", ",
namesof("shape.i", lshape.i, eshape.i, tag = FALSE)),
if (li.mix && ld.mix) ", \n ",
if (ld.mix > 0) c( if (li.mix) "" else ", ",
"multilogit(pdip.mix)"),
if (ld.mix > 1) c(", ",
namesof("shape.d", lshape.d, eshape.d, tag = FALSE)),
if (la.mlm) paste0(",\n",
paste0(" multilogit(", tmp7a, collapse = "),\n"),
")") else NULL,
if (li.mlm) paste0(",\n",
paste0(" multilogit(", tmp7b, collapse = "),\n"),
")") else NULL,
if (ld.mlm) paste0(",\n",
paste0(" multilogit(", tmp7c, collapse = "),\n"),
")") else NULL),
constraints = eval(substitute(expression({
M1 <- max(extra$indeta, na.rm = TRUE)
la.mix <- ( .la.mix )
li.mix <- ( .li.mix )
ld.mix <- ( .ld.mix )
la.mlm <- ( .la.mlm )
li.mlm <- ( .li.mlm )
ld.mlm <- ( .ld.mlm )
use.mat.mlm.a <- if (la.mlm) {
if ( .parallel.a ) matrix(1, la.mlm, 1) else diag(la.mlm)
} else {
NULL
}
use.mat.mlm.i <- if (li.mlm) {
if ( .parallel.i ) matrix(1, li.mlm, 1) else diag(li.mlm)
} else {
NULL
}
use.mat.mlm.d <- if (ld.mlm) {
if ( .parallel.d ) matrix(1, ld.mlm, 1) else diag(ld.mlm)
} else {
NULL
}
if (la.mlm + li.mlm + ld.mlm == 0) {
Use.mat <- use.mat.mlm <- cbind(M) # shape.p only
}
if (la.mlm + li.mlm + ld.mlm) {
nc1 <- if (length(use.mat.mlm.a)) ncol(use.mat.mlm.a) else 0
nc2 <- if (length(use.mat.mlm.i)) ncol(use.mat.mlm.i) else 0
nc3 <- if (length(use.mat.mlm.d)) ncol(use.mat.mlm.d) else 0
use.mat.mlm <- cbind(1, matrix(0, 1, nc1 + nc2 + nc3))
if (la.mlm)
use.mat.mlm <- rbind(use.mat.mlm,
cbind(matrix(0, la.mlm, 1),
use.mat.mlm.a,
if (length(use.mat.mlm.i) == 0)
NULL else matrix(0, la.mlm, nc2),
if (length(use.mat.mlm.d) == 0)
NULL else matrix(0, la.mlm, nc3)))
if (li.mlm )
use.mat.mlm <-
rbind(use.mat.mlm,
cbind(matrix(0, li.mlm, 1 + nc1),
use.mat.mlm.i,
matrix(0, li.mlm, nc3)))
if (ld.mlm)
use.mat.mlm <-
rbind(use.mat.mlm, # zz1 next line:
cbind(matrix(0, ld.mlm, 1 + nc1 + nc2),
use.mat.mlm.d))
} # la.mlm + li.mlm
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
use.mat.mix <- cm3gaitd( .eq.ap , .eq.ip , .eq.dp , npar = 1)
tmp3.subset <- tmp3.TF[-(8:10)]
use.mat.mix <- use.mat.mix[tmp3.subset, , drop = FALSE]
notall0 <- function(x) !all(x == 0)
use.mat.mix <- use.mat.mix[, apply(use.mat.mix, 2, notall0),
drop = FALSE]
if (la.mix + li.mix + ld.mix > 0)
Use.mat <- use.mat.mix
if (la.mlm + li.mlm + ld.mlm > 0) {
Use.mat <- rbind(use.mat.mix,
matrix(0, nrow(use.mat.mlm) - 1, # bottom
ncol(use.mat.mix)))
Use.mat <- cbind(Use.mat,
matrix(0, nrow(Use.mat), # RHS
ncol(use.mat.mlm) - 1))
Use.mat[row(Use.mat) > nrow(use.mat.mix) &
col(Use.mat) > ncol(use.mat.mix)] <- use.mat.mlm[-1, -1]
} # la.mlm + li.mlm + ld.mlm > 0
if (is.null(constraints)) {
constraints <-
cm.VGAM(Use.mat, x = x, apply.int = TRUE, # FALSE
bool = .eq.ap || .eq.ip || .eq.dp ||
.parallel.a || .parallel.i || .parallel.d ,
constraints = constraints) # FALSE
} # is.null(constraints)
if (la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm)
constraints <-
cm.zero.VGAM(constraints, x = x, .zero , M = M, M1 = M1,
predictors.names = paste0(predictors.names,
names(predictors.names)))
}),
list( .zero = zero, .tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.eq.ap = eq.ap, .eq.ip = eq.ip, .eq.dp = eq.dp,
.parallel.a = parallel.a, .parallel.i = parallel.i,
.parallel.d = parallel.d,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix ))),
infos = eval(substitute(function(...) {
list(M1 = .M1 ,
Q1 = 1,
dpqrfun = "gaitdlog",
link = .predictors.names , # ...strips... from above
link1parameter = as.logical( .lall.len <= 2), # <= 1 safer
mixture.links = any(c( .la.mlm , .li.mlm , .ld.mlm ,
.la.mix , .li.mix ,
.ld.mix ) > 1), # FALSE if NULL
a.mix = as.vector( .a.mix ), # Handles NULL
a.mlm = as.vector( .a.mlm ),
i.mix = as.vector( .i.mix ),
i.mlm = as.vector( .i.mlm ),
d.mix = as.vector( .d.mix ),
d.mlm = as.vector( .d.mlm ),
truncate = as.vector( .truncate ),
max.support = as.vector( .max.support ),
Support = c( .lowsup , Inf, 1), # a(b)c format as a,c,b.
expected = TRUE,
multipleResponses = FALSE, # poissonff can b called ifTRUE
parameters.names = names( .predictors.names ),
parent.name = c("logff", "log"),
type.fitted = as.vector( .type.fitted ),
type.fitted.choices = ( .type.fitted.choices ),
baseparams.argnames = "shape",
MM1 = 1, # One parameter for 1 response (shape). Needed.
zero = .zero )
}, list( .zero = zero, .lowsup = lowsup,
.type.fitted = type.fitted,
.type.fitted.choices = type.fitted.choices,
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix,
.truncate = truncate, .max.support = max.support,
.predictors.names = predictors.names,
.M1 = M1, .lall.len = lall.len
))),
rqresslot = eval(substitute(
function(mu, y, w, eta, extra = NULL) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least 1 shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vecs
ind.shape.z <- c(na.omit(ind.shape.z)) # At least 1 value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
scrambleseed <- runif(1) # To scramble the seed
qnorm(runif(length(y),
pgaitdlog(y - 1, shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm),
pgaitdlog(y , shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)))
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
initialize = eval(substitute(expression({
extra$indeta <- ( .indeta ) # Avoids recomputing it
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix +
la.mlm + li.mlm + ld.mlm
truncate <- as.vector( .truncate )
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(y)
M <- NOS * M1
tmp3.TF <- ( .tmp3.TF )
temp5 <-
w.y.check(w = w, y = y,
Is.nonnegative.y = TRUE,
Is.integer.y = TRUE,
ncol.w.max = 1, # Since max.support = 9 is possible
ncol.y.max = 1,
out.wy = TRUE, colsyperw = 1, maximize = TRUE)
w <- temp5$w
y <- temp5$y
glist <- y.gaitcombo.check(y, truncate = truncate,
a.mlm = a.mlm, a.mix = a.mix,
i.mlm = i.mlm, i.mix = i.mix,
d.mlm = d.mlm, d.mix = d.mix,
max.support = .max.support ,
min.support = .min.support )
extra$skip.mix.a <- glist$skip.mix.a
extra$skip.mix.i <- glist$skip.mix.i
extra$skip.mix.d <- glist$skip.mix.d
extra$skip.mlm.a <- glist$skip.mlm.a
extra$skip.mlm.i <- glist$skip.mlm.i
extra$skip.mlm.d <- glist$skip.mlm.d
extra$NOS <- NOS <- ncoly <- ncol(y) # Number of species
extra$type.fitted <- as.vector( .type.fitted )
extra$mux.init <- as.vector( .mux.init )
extra$colnames.y <- colnames(y)
extra$M1 <- M1
extra$index.M <- iam(NA, NA, M, both = TRUE) # Used in @weight
predictors.names <- ( .predictors.names ) # Got it, named
if (!length(etastart)) {
shape.p.init <- if (length( .ishape.p )) .ishape.p else {
logff.Loglikfun <- function(shapeval, y, x, w, extraargs) {
sum(c(w) * dlog(x = y, shape = shapeval, log = TRUE))
}
shape.p.grid <- ( .gshape.p )
grid.search(shape.p.grid,
objfun = logff.Loglikfun, y = y, w = w)
}
shape.p.init <- rep(shape.p.init, length = n)
shape.d.init <-
shape.a.init <- shape.i.init <- shape.p.init # Needed
etastart <- matrix(nrow = n, ncol = M,
theta2eta(shape.p.init, .lshape.p , earg = .eshape.p ))
mux.more.a <- extra$mux.init[1] # 0.75 Err 2 bit smaller
init.pobs.mix <- numeric(n)
if (tmp3.TF[ 2]) { # la.mix > 0
init.pobs.mix <- if (length( .ipobs.mix )) {
rep_len( .ipobs.mix , n)
} else {
is.a.mix1 <- rowSums(extra$skip.mix.a) > 0
rep(mux.more.a * sum(w[is.a.mix1]) / sum(w), n)
}
} # la.mix > 0
if (tmp3.TF[ 3]) { # Assign coln 3; la.mix > 1
shape.a.init <- if (length( .ishape.a ))
rep_len( .ishape.a , n) else shape.p.init # A vector
etastart[, 3] <-
theta2eta(shape.a.init, .lshape.a , earg = .eshape.a )
}
init.pstr.mix <- init.pdip.mix <- numeric(n)
try.gridsearch.pstr.mix <- FALSE
if (tmp3.TF[ 4]) { # li.mix > 0
init.pstr.mix <- if (length( .ipstr.mix )) {
rep_len( .ipstr.mix , n)
} else {
try.gridsearch.pstr.mix <- TRUE
numeric(n) # Overwritten by gridsearch
}
} # li.mix > 0
if (tmp3.TF[ 5]) { # li.mix > 1
shape.i.init <- if (length( .ishape.i ))
rep_len( .ishape.i , n) else shape.p.init # A vector
etastart[, (extra$indeta[5, 'launch'])] <-
theta2eta(shape.i.init, .lshape.i , earg = .eshape.i )
} # li.mix > 1
if (tmp3.TF[ 8]) { # la.mlm
init.pobs.mlm <- if (length( .ipobs.mlm )) {
matrix( .ipobs.mlm , n, la.mlm, byrow = .byrow.aid )
} else {
mux.more.a <- extra$mux.init[1]
init.pobs.mlm <- colSums(c(w) *
extra$skip.mlm.a) / colSums(w)
init.pobs.mlm <- init.pobs.mlm * as.vector( mux.more.a )
matrix(init.pobs.mlm, n, la.mlm, byrow = TRUE)
}
} else {
init.pobs.mlm <- matrix(0, n, 1)
}
try.gridsearch.pstr.mlm <- FALSE
if (tmp3.TF[ 9]) { # li.mlm
try.gridsearch.pstr.mlm <- !(length( .ipstr.mlm ))
init.pstr.mlm <- 0 # Might be overwritten by gridsearch
if (length( .ipstr.mlm ))
init.pstr.mlm <- as.vector( .ipstr.mlm )
init.pstr.mlm <- matrix(init.pstr.mlm, n, li.mlm,
byrow = .byrow.aid )
} else {
init.pstr.mlm <- matrix(0, n, 1)
}
init.pdip.mlm <- matrix(0, n, 2) # rowSums() needs > 1 colns.
gaitdlog.Loglikfun1.mix <-
function(pstr.mix.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mix = pstr.mix.val,
pstr.mlm = extraargs$pstr.mlm, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdlog.Loglikfun1.mlm <-
function(pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mlm = pstr.mlm.val,
pstr.mix = extraargs$pstr.mix, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdlog.Loglikfun2 <-
function(pstr.mix.val, pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mix = pstr.mix.val,
pstr.mlm = pstr.mlm.val,
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
if (li.mix + li.mlm) {
extraargs <- list(
shape.p = shape.p.init,
shape.a = shape.a.init,
shape.i = shape.i.init,
shape.d = shape.d.init,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
pobs.mix = init.pobs.mix ,
pobs.mlm = init.pobs.mlm ,
pdip.mix = init.pdip.mix ,
pdip.mlm = init.pdip.mlm )
pre.warn <- options()$warn
options(warn = -1) # Ignore warnings during gridsearch
try.this <-
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
grid.search2( .gpstr.mix , .gpstr.mlm ,
objfun = gaitdlog.Loglikfun2,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mix) {
extraargs$pstr.mlm <- init.pstr.mlm
grid.search ( .gpstr.mix ,
objfun = gaitdlog.Loglikfun1.mix,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mlm) {
extraargs$pstr.mix <- init.pstr.mix
grid.search ( .gpstr.mlm ,
objfun = gaitdlog.Loglikfun1.mlm,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
}
options(warn = pre.warn) # Restore warnings
if (any(is.na(try.this)))
warning("gridsearch returned NAs. It's going to crash.",
immediate. = TRUE)
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
init.pstr.mix <- rep_len(try.this["Value1"], n)
init.pstr.mlm <- matrix(try.this["Value2"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'",
" and/or 'gpstr.mlm = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mix) {
init.pstr.mix <- rep_len(try.this["Value"], n)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mlm) {
init.pstr.mlm <- matrix(try.this["Value"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mlm = seq(5) / 100'.")
}
} # la.mix + lnf.mix
mux.more.d <- extra$mux.init[3]
if (ld.mix) {
init.pdip.mix <- if (length( .ipdip.mix ))
rep_len( .ipdip.mix, n) else {
is.d.mix1 <- rowSums(extra$skip.mix.d) > 0
rep(mux.more.d * sum(w[is.d.mix1]) / sum(w), n)
}
} # ld.mix
if (ld.mlm) {
init.pdip.mlm <- if (length( .ipdip.mlm ))
matrix( .ipdip.mlm, n, ld.mlm, byrow = TRUE) else {
is.d.mlm1 <- rowSums(extra$skip.mlm.d) > 0
matrix(mux.more.d * (sum(w[is.d.mlm1]) / sum(w)) / ld.mlm,
n, ld.mlm)
}
} # ld.mlm
while (any((vecTF <- init.pobs.mix + init.pstr.mix + # -
init.pdip.mix +
rowSums(init.pobs.mlm) +
rowSums(init.pstr.mlm) + # -
rowSums(init.pdip.mlm) > 0.96875))) {
init.pobs.mix[vecTF] <- 0.875 * init.pobs.mix[vecTF]
init.pstr.mix[vecTF] <- 0.875 * init.pstr.mix[vecTF]
init.pdip.mix[vecTF] <- 0.875 * init.pdip.mix[vecTF]
init.pobs.mlm[vecTF, ] <- 0.875 * init.pobs.mlm[vecTF, ]
init.pstr.mlm[vecTF, ] <- 0.875 * init.pstr.mlm[vecTF, ]
init.pdip.mlm[vecTF, ] <- 0.875 * init.pdip.mlm[vecTF, ]
} # while
Numer.init1 <- 1 - rowSums(init.pobs.mlm) -
rowSums(init.pstr.mlm) - # +
rowSums(init.pdip.mlm) -
init.pobs.mix - init.pstr.mix - # +
init.pdip.mix # Differs from 'Numer'.
etastart.z <- if (lall.len == 0) NULL else {
tmp.mat <- cbind(if (tmp3.TF[ 2]) init.pobs.mix else NULL,
if (tmp3.TF[ 4]) init.pstr.mix else NULL,
if (tmp3.TF[ 6]) init.pdip.mix else NULL,
if (tmp3.TF[ 8]) init.pobs.mlm else NULL,
if (tmp3.TF[ 9]) init.pstr.mlm else NULL,
if (tmp3.TF[10]) init.pdip.mlm else NULL,
Numer.init1)
multilogitlink(tmp.mat)
} # etastart.z
if (!is.matrix(etastart.z)) etastart.z <- cbind(etastart.z)
nextone <- 1 # Might not be used actually
if (tmp3.TF[ 2]) {
etastart[, 2] <- etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 4]) { # Coln 2 or 4
etastart[, (extra$indeta[4, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 6]) { # Coln 2 or 4 or 6
etastart[, (extra$indeta[6, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 8]) {
ind8 <- (extra$indeta[8, 'launch']):(
extra$indeta[8, 'finish'])
etastart[, ind8] <- etastart.z[, nextone:(nextone+
la.mlm - 1)]
nextone <- nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (extra$indeta[9, 'launch']):(
extra$indeta[9, 'finish'])
etastart[, ind9] <- etastart.z[, nextone:(nextone+
li.mlm - 1)]
nextone <- nextone + li.mlm
}
if (tmp3.TF[10]) {
ind0 <- (extra$indeta[10, 'launch']):(
extra$indeta[10, 'finish'])
etastart[, ind0] <- etastart.z[, nextone:(nextone +
ld.mlm - 1)]
if (ncol(etastart.z) != nextone + ld.mlm - 1)
stop("miscalculation")
}
}
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.ishape.p = ishape.p,
.ishape.a = ishape.a,
.ishape.i = ishape.i,
.ishape.d = ishape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.lpdip.mix = lpdip.mix,
.epdip.mix = epdip.mix,
.ipstr.mix = ipstr.mix, .ipobs.mix = ipobs.mix,
.ipstr.mlm = ipstr.mlm, .ipobs.mlm = ipobs.mlm,
.ipdip.mix = ipdip.mix,
.ipdip.mlm = ipdip.mlm,
.byrow.aid = byrow.aid,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support,
.min.support = lowsup,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.predictors.names = predictors.names,
.mux.init = mux.init,
.gshape.p = gshape.p,
.gpstr.mix = gpstr.mix, # .gpdip.mix = gpdip.mix,
.gpstr.mlm = gpstr.mlm, # .gpdip.mlm = gpdip.mlm,
.ishrinkage = ishrinkage, .probs.y = probs.y,
.indeta = indeta,
.imethod = imethod, .type.fitted = type.fitted ))),
linkinv = eval(substitute(function(eta, extra = NULL) {
n.obs <- NROW(eta)
type.fitted <-
if (length(extra$type.fitted)) extra$type.fitted else {
warning("cannot find 'type.fitted'. Returning the 'mean'.")
"mean"
}
type.fitted <-
match.arg(type.fitted[1],
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"))[1]
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
morework <- type.fitted != "mean" # For efficiency
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(eta) / M1
Bits <- moments.gaitdcombo.log(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
if (morework) {
Denom.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p == 0)) {
smallval <- min(Denom.p[Denom.p > 0])
Denom.p[Denom.p == 0] <- 1e-09 # smallval
warning("0s found in variable 'Denom.p'. Trying to fix it.")
}
Numer <- c(1 - pobs.mix - pstr.mix + pdip.mix -
(if (la.mlm) rowSums(pobs.mlm) else 0) -
(if (li.mlm) rowSums(pstr.mlm) else 0) +
(if (ld.mlm) rowSums(pdip.mlm) else 0))
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom.p)
} # morework
if (!la.mlm && type.fitted %in% c("pobs.mlm")) {
warning("No altered MLM values; returning an NA")
return(NA)
}
if (!li.mlm && type.fitted %in% c("sum.mlm.i", "pstr.mlm")) {
warning("No inflated MLM values; returning an NA")
return(NA)
}
if (!ld.mlm && type.fitted %in% c("sum.mlm.d", "pdip.mlm")) {
warning("No deflated MLM values; returning an NA")
return(NA)
}
if (!la.mix && type.fitted %in% c("Pobs.mix")) {
warning("No altered mixture values; returning an NA")
return(NA)
}
if (!li.mix && type.fitted %in% c("sum.mix.i", "Pstr.mix")) {
warning("No inflated mixture values; returning an NA")
return(NA)
}
if (!ld.mix && type.fitted %in% c("sum.mix.d", "Pdip.mix")) {
warning("No deflated mixture values; returning an NA")
return(NA)
}
if (la.mix && morework) {
tmp13 <- # dpois() does not retain the matrix format
dlog(matrix(a.mix, n.obs, la.mix, byrow = TRUE),
matrix(shape.a, n.obs, la.mix)) / (
c(Bits[["SumA0.mix.a"]]))
dim(tmp13) <- c(n.obs, la.mix)
dimnames(tmp13) <- list(rownames(eta),
as.character(a.mix))
propn.mat.a <- tmp13
} # la.mix
if (li.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dlog(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.i, n.obs, li.mix)) / (
c(Bits[["SumI0.mix.i"]]))
dim(tmp55) <- c(n.obs, li.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(i.mix))
propn.mat.i <- tmp55 # Correct dimension
} # li.mix
if (ld.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dlog(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.d, n.obs, ld.mix)) / (
c(Bits[["SumD0.mix.d"]]))
dim(tmp55) <- c(n.obs, ld.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(d.mix))
propn.mat.d <- tmp55 # Correct dimension
} # ld.mix
ans <- switch(type.fitted,
"mean" = Bits[["mean"]], # Unconditional mean
"shapes" = cbind(shape.p,
if (tmp3.TF[ 3]) shape.a else NULL,
if (tmp3.TF[ 5]) shape.i else NULL,
if (tmp3.TF[ 7]) shape.d else NULL),
"pobs.mlm" = pobs.mlm, # aka omegamat, n x la.mlm
"pstr.mlm" = pstr.mlm, # aka phimat, n x li.mlm
"pdip.mlm" = pdip.mlm, # aka psimat, n x ld.mlm
"pobs.mix" = pobs.mix, # n-vector
"pstr.mix" = pstr.mix, # n-vector
"pdip.mix" = pdip.mix, # n-vector
"Pobs.mix" = c(pobs.mix) * propn.mat.a, # matrix
"Pstr.mix" = c(pstr.mix) * propn.mat.i,
"Pdip.mix" = c(pdip.mix) * propn.mat.d,
"nonspecial" = probns,
"Numer" = Numer,
"Denom.p" = Denom.p,
"sum.mlm.i" = pstr.mlm + Numer *
dlog(matrix(i.mlm, n.obs, li.mlm, byrow = TRUE),
matrix(shape.p, n.obs, li.mlm)) / Denom.p,
"sum.mlm.d" = -pdip.mlm + Numer *
dlog(matrix(d.mlm, n.obs, ld.mlm, byrow = TRUE),
matrix(shape.p, n.obs, ld.mlm)) / Denom.p,
"sum.mix.i" = c(pstr.mix) * propn.mat.i + Numer *
dlog(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.p, n.obs, li.mix)) / Denom.p,
"sum.mix.d" = -c(pdip.mix) * propn.mat.d + Numer *
dlog(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.p, n.obs, ld.mix)) / Denom.p,
"ptrunc.p" = Bits[["SumT0.p"]] + 1 - Bits[["cdf.max.s"]],
"cdf.max.s" = Bits[["cdf.max.s"]]) # Pr(y <= max.support)
ynames.pobs.mlm <- as.character(a.mlm) # Works with NULLs
ynames.pstr.mlm <- as.character(i.mlm) # Works with NULLs
ynames.pdip.mlm <- as.character(d.mlm) # Works with NULLs
if (length(ans))
label.cols.y(ans, NOS = NOS, colnames.y =
switch(type.fitted,
"shapes" = c("shape.p", "shape.a", # Some colns NA
"shape.i", "shape.d")[(tmp3.TF[c(1, 3, 5, 7)])],
"Pobs.mix" = as.character(a.mix),
"sum.mix.i" = , #
"Pstr.mix" = as.character(i.mix),
"sum.mix.d" = , #
"Pdip.mix" = as.character(d.mix),
"pobs.mlm" = ynames.pobs.mlm,
"sum.mlm.i" = , #
"pstr.mlm" = ynames.pstr.mlm,
"sum.mlm.d" = , #
"pdip.mlm" = ynames.pdip.mlm,
extra$colnames.y)) else ans
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
last = eval(substitute(expression({
pred.names <- c( .predictors.names ) # Save it
link.names <- as.vector( .predictors.names )
parameter.names <- names(pred.names)
predictors.names <- NULL
for (jay in seq(M))
predictors.names <- c(predictors.names,
namesof(parameter.names[jay], link.names[jay], tag = FALSE,
earg = list())) # This isnt perfect; info is lost
misc$predictors.names <- predictors.names # Useful for coef()
misc$link <- link.names #
names(misc$link) <- parameter.names #
misc$earg <- vector("list", M1)
names(misc$earg) <- names(misc$link)
misc$earg[[1]] <- ( .eshape.p ) # First one always there
iptr <- 1
if (tmp3.TF[ 2])
misc$earg[[(iptr <- iptr + 1)]] <- list() # multilogitlink
if (tmp3.TF[ 3])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.a )
if (tmp3.TF[ 4])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 5])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.i )
if (tmp3.TF[ 6])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 7])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.d )
if (tmp3.TF[ 8]) { # la.mlm
for (ii in seq(la.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # la.mlm
if (tmp3.TF[ 9]) { # li.mlm
for (ii in seq(li.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # li.mlm
if (tmp3.TF[10]) { # ld.mlm
for (ii in seq(ld.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # ld.mlm
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.predictors.names = predictors.names,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
loglikelihood = eval(substitute(
function(mu, y, w, residuals = FALSE, eta,
extra = NULL, summation = TRUE) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
if (residuals) {
stop("loglikelihood residuals not implemented yet")
} else {
ll.elts <- c(w) *
dgaitdlog(y, shape.p, log = TRUE, # byrow.aid = F,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)
if (summation) {
sum(ll.elts)
} else {
ll.elts
}
}
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
vfamily = c("gaitdlog"),
validparams = eval(substitute(function(eta, y, extra = NULL) {
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
small. <- 1e-14
pobs.mix <- pstr.mix <- pdip.mix <- small. # 4 rowSums():
pobs.mlm <- pstr.mlm <- pdip.mlm <- matrix(small., NROW(eta), 1)
shape.a <- shape.i <- shape.d <- 0.5 # Needed
if (!is.matrix(eta)) eta <- as.matrix(eta)
shape.p <-
cbind(eta2theta(eta[, 1], .lshape.p , earg = .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 1] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
okay.mlm <-
all(is.finite(pobs.mlm)) && all(0 < pobs.mlm) &&
all(is.finite(pstr.mlm)) && all(0 < pstr.mlm) &&
all(is.finite(pdip.mlm)) && all(0 < pdip.mlm)
okay.mix <-
all(is.finite(shape.p)) && all(0 < shape.p) && all(shape.p < 1) &&
all(is.finite(shape.a)) && all(0 < shape.a) && all(shape.a < 1) &&
all(is.finite(shape.i)) && all(0 < shape.i) && all(shape.i < 1) &&
all(is.finite(shape.d)) && all(0 < shape.d) && all(shape.d < 1) &&
all(is.finite(pobs.mix)) && all(0 < pobs.mix) &&
all(is.finite(pstr.mix)) && all(0 < pstr.mix) &&
all(is.finite(pdip.mix)) && all(0 < pdip.mix) &&
all(pobs.mix + pstr.mix + pdip.mix +
rowSums(pobs.mlm) + rowSums(pstr.mlm) +
rowSums(pdip.mlm) < 1) # Combined
okay.mlm && okay.mix
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
simslot = eval(substitute(
function(object, nsim) {
pwts <- if (length(pwts <- object@prior.weights) > 0)
pwts else weights(object, type = "prior")
if (any(pwts != 1))
warning("ignoring prior weights")
eta <- predict(object)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
extra <- object@extra
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF )
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A AMLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
rgaitdlog(nsim * length(shape.p), shape.p,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = .truncate , max.support = .max.support )
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
deriv = eval(substitute(expression({
tmp3.TF <- ( .tmp3.TF )
calA.p <- tmp3.TF[ 2]
calI.p <- tmp3.TF[ 4]
calD.p <- tmp3.TF[ 6]
calA.np <- tmp3.TF[ 8]
calI.np <- tmp3.TF[ 9]
calD.np <- tmp3.TF[10]
Denom1.a <- Denom1.i <- Denom1.d <-
Denom2.i <- Denom2.d <- 0 # Denom2.a is unneeded
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted.
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
minprob.baseline <- min(allprobs[, ncol(allprobs)], na.rm = TRUE)
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1) {
warning("fitted probabilities numerically 0 or 1 occurred")
} else
if (minprob.baseline < 0.10)
warning("Minimum baseline (reserve) probability close to 0")
if (control$trace)
cat("Minimum baseline (reserve) probability = ",
format(minprob.baseline, digits = 3), "\n")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- ncol(eta) / M1 # extra$NOS
if (NOS != 1) stop("can only handle 1 response")
is.a.mixed <- if (tmp3.TF[ 2])
rowSums(extra$skip.mix.a) > 0 else rep(FALSE, n)
is.i.mixed <- if (tmp3.TF[ 4])
rowSums(extra$skip.mix.i) > 0 else rep(FALSE, n)
is.d.mixed <- if (tmp3.TF[ 6])
rowSums(extra$skip.mix.d) > 0 else rep(FALSE, n)
is.a.mlmed <- if (tmp3.TF[ 8])
rowSums(extra$skip.mlm.a) > 0 else rep(FALSE, n)
is.i.mlmed <- if (tmp3.TF[ 9])
rowSums(extra$skip.mlm.i) > 0 else rep(FALSE, n)
is.d.mlmed <- if (tmp3.TF[10])
rowSums(extra$skip.mlm.d) > 0 else rep(FALSE, n)
is.ns <- !is.a.mlmed & !is.i.mlmed & !is.d.mlmed &
!is.a.mixed & !is.i.mixed & !is.d.mixed # & !is.truncd
A8.p <- -1 / log1p(-shape.p)
A8.a <- -1 / log1p(-shape.a)
A8.i <- -1 / log1p(-shape.i)
prob.mlm.a <- if (la.mlm) rowSums(pobs.mlm) else 0 # scalar okay
prob.mlm.i <- if (li.mlm) rowSums(pstr.mlm) else 0 # scalar okay
prob.mlm.d <- if (ld.mlm) rowSums(pdip.mlm) else 0 # scalar okay
pmf.deriv1 <- function(y, shape) {
A8 <- -1 / log1p(-shape)
deriv0 <- A8 * (shape^y) / y
A8 * (shape^(y-1) - deriv0 / (1 - shape))
}
pmf.deriv2 <- function(y, shape) {
A8 <- -1 / log1p(-shape)
A8prime <- -(A8^2) / (1 - shape)
deriv0 <- A8 * (shape^y) / y
deriv1 <- A8 * (shape^(y-1) - deriv0 / (1 - shape))
A8prime * (shape^(y-1) - deriv0 / (1 - shape)) +
A8 * ((y - 1) * shape^(y - 2) - deriv0 / (1 - shape)^2 -
deriv1 / (1 - shape))
}
sumD.mix.1a.p <- sumD.mix.2a.p <- matrix(0, n, NOS)
if (la.mix > 0) { # \calA_p
DA.mix.0mat.a <- # Matches naming convention further below
DA.mix.1mat.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
sumD.mix.1a.p <- sumD.mix.1a.p + pmf.deriv1(aval, shape.p)
sumD.mix.2a.p <- sumD.mix.2a.p + pmf.deriv2(aval, shape.p)
pmf.a <- dlog(aval, shape.a)
DA.mix.0mat.a[, jay] <- pmf.a
DA.mix.1mat.a[, jay] <- pmf.deriv1(aval, shape.a)
}
Denom1.a <- rowSums(DA.mix.1mat.a) # aka sumD.mix.1a.a
} # la.mix > 0
if (li.mix) {
DI.mix.0mat.i <- # wrt inflated distribution
DI.mix.1mat.i <- DI.mix.2mat.i <- matrix(0, n, li.mix)
DP.mix.0mat.i <- # wrt parent distribution
DP.mix.1mat.i <- DP.mix.2mat.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.i <- dlog(ival, shape.i)
DI.mix.0mat.i[, jay] <- pmf.i
DI.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.i)
DI.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.i)
pmf.p <- dlog(ival, shape.p)
DP.mix.0mat.i[, jay] <- pmf.p
DP.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.p)
DP.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
Denom1.i <- rowSums(DI.mix.1mat.i)
Denom2.i <- rowSums(DI.mix.2mat.i)
} # li.mix
if (ld.mix) {
DD.mix.0mat.d <- # wrt deflated distribution
DD.mix.1mat.d <- DD.mix.2mat.d <- matrix(0, n, ld.mix)
DP.mix.0mat.d <- # wrt parent distribution
DP.mix.1mat.d <- DP.mix.2mat.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.d <- dlog(dval, shape.d)
DD.mix.0mat.d[, jay] <- pmf.d
DD.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.d)
DD.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.d)
pmf.p <- dlog(dval, shape.p)
DP.mix.0mat.d[, jay] <- pmf.p
DP.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.p)
DP.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
Denom1.d <- rowSums(DD.mix.1mat.d)
Denom2.d <- rowSums(DD.mix.2mat.d)
} # ld.mix
Bits <- moments.gaitdcombo.log(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
sumD.mlm.1a.p <- sumD.mlm.2a.p <- matrix(0, n, NOS)
if (la.mlm)
for (aval in a.mlm) {
sumD.mlm.1a.p <- sumD.mlm.1a.p + pmf.deriv1(aval, shape.p)
sumD.mlm.2a.p <- sumD.mlm.2a.p + pmf.deriv2(aval, shape.p)
}
Denom0.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
Numer <- 1 - pobs.mix - pstr.mix - prob.mlm.a - prob.mlm.i +
pdip.mix + prob.mlm.d
Denom0.a <- c(Bits[["SumA0.mix.a"]]) # Not .p
Denom0.i <- c(Bits[["SumI0.mix.i"]])
Denom0.d <- c(Bits[["SumD0.mix.d"]])
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, NOS)
if (li.mlm > 0) {
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
pmf.p <- dlog(ival, shape.p)
Dp.mlm.0Mat.i[, jay] <- pmf.p
Dp.mlm.1Mat.i[, jay] <- pmf.deriv1(ival, shape.p)
Dp.mlm.2Mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
} # li.mlm
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, NOS)
if (ld.mlm > 0) {
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
pmf.p <- dlog(dval, shape.p)
Dp.mlm.0Mat.d[, jay] <- pmf.p
Dp.mlm.1Mat.d[, jay] <- pmf.deriv1(dval, shape.p)
Dp.mlm.2Mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
} # ld.mlm
sumD.1t.p <- sumD.2t.p <-
sumD.1t.a <- sumD.2t.a <-
sumD.1t.i <- sumD.2t.i <-
sumD.1t.d <- sumD.2t.d <- matrix(0, n, NOS)
if (ltruncat)
for (tval in truncate) {
sumD.1t.p <- sumD.1t.p + pmf.deriv1(tval, shape.p)
sumD.2t.p <- sumD.2t.p + pmf.deriv2(tval, shape.p)
sumD.1t.a <- sumD.1t.a + pmf.deriv1(tval, shape.a)
sumD.2t.a <- sumD.2t.a + pmf.deriv2(tval, shape.a)
sumD.1t.i <- sumD.1t.i + pmf.deriv1(tval, shape.i)
sumD.2t.i <- sumD.2t.i + pmf.deriv2(tval, shape.i)
sumD.1t.d <- sumD.1t.d + pmf.deriv1(tval, shape.d)
sumD.2t.d <- sumD.2t.d + pmf.deriv2(tval, shape.d)
}
if (is.finite(max.support)) {
tmp1.p <- A8.p * (shape.p^max.support -
(1 - plog(max.support, shape.p))) / (1 - shape.p)
sumD.1t.p <- sumD.1t.p + tmp1.p
sumD.2t.p <- sumD.2t.p + (A8.p / (1 - shape.p)) * (
(shape.p^max.support) / (1 - shape.p) +
max.support * shape.p^(max.support - 1) -
(1 - plog(max.support, shape.p)) / (1 - shape.p) - 2 * tmp1.p)
tmp1.a <- A8.a * (shape.a^max.support -
(1 - plog(max.support, shape.a))) / (1 - shape.a)
sumD.1t.a <- sumD.1t.a + tmp1.a
sumD.2t.a <- sumD.2t.a + (A8.a / (1 - shape.a)) * (
(shape.a^max.support) / (1 - shape.a) +
max.support * shape.a^(max.support - 1) -
(1 - plog(max.support, shape.a)) / (1 - shape.a) - 2 * tmp1.a)
tmp1.i <- A8.i * (shape.i^max.support -
(1 - plog(max.support, shape.i))) / (1 - shape.i)
sumD.1t.i <- sumD.1t.i + tmp1.i
sumD.2t.i <- sumD.2t.i + (A8.i / (1 - shape.i)) * (
(shape.i^max.support) / (1 - shape.i) +
max.support * shape.i^(max.support - 1) -
(1 - plog(max.support, shape.i)) / (1 - shape.i) - 2 * tmp1.i)
} # is.finite(max.support)
Denom1.p <- c(-sumD.1t.p - sumD.mlm.1a.p - sumD.mix.1a.p)
Denom2.p <- c(-sumD.2t.p - sumD.mlm.2a.p - sumD.mix.2a.p)
d0B.PI.mlm <- Dp.mlm.0Mat.i / Denom0.p
d1B.PI.mlm <- Dp.mlm.1Mat.i / Denom0.p - # This is most general
Dp.mlm.0Mat.i * Denom1.p / Denom0.p^2
d2B.PI.mlm <- Dp.mlm.2Mat.i / Denom0.p -
2 * Dp.mlm.1Mat.i * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.i * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.i * (Denom1.p^2) / Denom0.p^3
d0B.PD.mlm <- Dp.mlm.0Mat.d / Denom0.p
d1B.PD.mlm <- Dp.mlm.1Mat.d / Denom0.p - # This is most general
Dp.mlm.0Mat.d * Denom1.p / Denom0.p^2
d2B.PD.mlm <- Dp.mlm.2Mat.d / Denom0.p -
2 * Dp.mlm.1Mat.d * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.d * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.d * (Denom1.p^2) / Denom0.p^3
DELTA.i.mlm <- if (li.mlm > 0) {
Numer * d0B.PI.mlm + pstr.mlm # n x li.mlm.
} else {
matrix(0, n, 1) # If li.mlm == 0, for rowSums().
}
DELTA.d.mlm <- if (ld.mlm > 0) {
Numer * d0B.PD.mlm - pdip.mlm # n x ld.mlm.
} else {
matrix(0, n, 1) # If ld.mlm == 0, for rowSums().
}
if (li.mix > 0) {
d0A.i <- DI.mix.0mat.i / Denom0.i
d0B.PI.mix <- DP.mix.0mat.i / Denom0.p
DELTA.i.mix <- Numer * d0B.PI.mix + pstr.mix * d0A.i
d1A.i <- (DI.mix.1mat.i - DI.mix.0mat.i *
Denom1.i / Denom0.i) / Denom0.i
d2A.i <- (DI.mix.2mat.i - (2 * DI.mix.1mat.i * Denom1.i +
DI.mix.0mat.i * Denom2.i) / Denom0.i +
2 * DI.mix.0mat.i * (Denom1.i / Denom0.i)^2) / Denom0.i
d1B.PI.mix <- DP.mix.1mat.i / Denom0.p -
DP.mix.0mat.i * Denom1.p / Denom0.p^2
d2B.PI.mix <- DP.mix.2mat.i / Denom0.p -
2 * DP.mix.1mat.i * Denom1.p / Denom0.p^2 -
DP.mix.0mat.i * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.i * (Denom1.p^2) / Denom0.p^3
} # li.mix > 0
if (ld.mix > 0) {
d0A.d <- DD.mix.0mat.d / Denom0.d
d0B.PD.mix <- DP.mix.0mat.d / Denom0.p
DELTA.d.mix <- Numer * d0B.PD.mix - pdip.mix * d0A.d
d1A.d <- (DD.mix.1mat.d - DD.mix.0mat.d *
Denom1.d / Denom0.d) / Denom0.d
d2A.d <- (DD.mix.2mat.d - (2 * DD.mix.1mat.d * Denom1.d +
DD.mix.0mat.d * Denom2.d) / Denom0.d +
2 * DD.mix.0mat.d * (Denom1.d / Denom0.d)^2) / Denom0.d
d1B.PD.mix <- DP.mix.1mat.d / Denom0.p -
DP.mix.0mat.d * Denom1.p / Denom0.p^2
d2B.PD.mix <- DP.mix.2mat.d / Denom0.p -
2 * DP.mix.1mat.d * Denom1.p / Denom0.p^2 -
DP.mix.0mat.d * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.d * (Denom1.p^2) / Denom0.p^3
} # ld.mix > 0
if (la.mix) {
d0A.a <- DA.mix.0mat.a / Denom0.a
d1A.a <- DA.mix.1mat.a / Denom0.a -
DA.mix.0mat.a * Denom1.a / Denom0.a^2
} # la.mix
dl.dshape.p <- -A8.p / (1 - shape.p) + y / shape.p
dl.dshape.p[!is.ns] <- 0 # For is.a.mixed & is.a.mlmed
dl.dshape.a <- dl.dshape.i <- dl.dshape.d <- numeric(n)
dl.dpstr.mix <- (-1) / Numer # \notin A, I, T, D
dl.dpstr.mix[is.a.mixed] <- 0
dl.dpstr.mix[is.a.mlmed] <- 0
dl.dpdip.mix <- (+1) / Numer # \notin A, I, T, D
dl.dpdip.mix[is.a.mixed] <- 0
dl.dpdip.mix[is.a.mlmed] <- 0
dl.dpobs.mix <- numeric(n) # 0 for \calA_{np}
dl.dpobs.mix[is.ns] <- (-1) / Numer[is.ns]
dl.dpobs.mlm <-
dl.dpstr.mlm <- matrix(0, n, 1) # May be unneeded
dl.dpdip.mlm <- matrix(0, n, max(1, ld.mlm)) # Initzed if used.
dl.dpdip.mlm[is.ns, ] <- 1 / Numer[is.ns]
if (tmp3.TF[ 8] && la.mlm) { # aka \calA_{np}
dl.dpobs.mlm <- matrix(-1 / Numer, n, la.mlm) # \notin calS
dl.dpobs.mlm[!is.ns, ] <- 0 # For a.mix only really
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
is.alt.j.mlm <- extra$skip.mlm.a[, jay] # Logical vector
tmp7a <- 1 / pobs.mlm[is.alt.j.mlm, jay]
dl.dpobs.mlm[is.alt.j.mlm, jay] <- tmp7a
} # jay
} # la.mlm
dl.dshape.p[is.ns] <- dl.dshape.p[is.ns] -
(Denom1.p / Denom0.p)[is.ns]
if (tmp3.TF[ 9] && li.mlm > 0) { # aka \calI_{np}
dl.dpstr.mlm <- matrix(-1 / Numer, n, li.mlm)
dl.dpstr.mlm[!is.ns, ] <- 0 # For a.mlm and a.mix
for (jay in seq(li.mlm)) {
is.inf.j.mlm <- extra$skip.mlm.i[, jay] # Logical vector
tmp7i <- Numer * d1B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
dl.dshape.p[is.inf.j.mlm] <- tmp7i[is.inf.j.mlm]
tmp9i <- d0B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
n.tmp <- -tmp9i[is.inf.j.mlm]
p.tmp <- +tmp9i[is.inf.j.mlm]
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mlm, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mlm ] <- p.tmp
tmp8 <- (1 - d0B.PI.mlm[, jay]) / DELTA.i.mlm[, jay]
dl.dpstr.mlm[is.inf.j.mlm, ] <- n.tmp # tmp9[is.inf.j.mlm]
dl.dpstr.mlm[is.inf.j.mlm, jay] <- tmp8[is.inf.j.mlm]
} # jay
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # aka \calD_{np}
for (jay in seq(ld.mlm)) {
is.def.j.mlm <- extra$skip.mlm.d[, jay] # Logical vector
tmp7d <- Numer * d1B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
dl.dshape.p[is.def.j.mlm] <- tmp7d[is.def.j.mlm] # 20211020
tmp9d <- d0B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
p.tmp <- +tmp9d[is.def.j.mlm]
n.tmp <- -tmp9d[is.def.j.mlm]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.def.j.mlm ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, jay] <-
dl.dpdip.mlm[is.def.j.mlm, jay] -
1 / DELTA.d.mlm[is.def.j.mlm, jay]
} # jay
} # ld.mlm > 0
if (tmp3.TF[ 2] && la.mix) { # aka \calA_{p}
dl.dpobs.mix[is.a.mixed] <- 1 / pobs.mix[is.a.mixed]
if (tmp3.TF[ 3] && la.mix > 1)
for (jay in seq(la.mix)) {
is.alt.j.mix <- extra$skip.mix.a[, jay] # Logical vector
tmp2 <- d1A.a[, jay] / d0A.a[, jay]
dl.dshape.a[is.alt.j.mix] <- tmp2[is.alt.j.mix] # ccc.
} # jay
} # la.mix
if (tmp3.TF[ 4] && li.mix > 0) { # aka \calI_{p}
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
is.inf.j.mix <- extra$skip.mix.i[, jay] # Logical vector
tmp7b <- Numer * d1B.PI.mix[, jay] / DELTA.i.mix[, jay]
dl.dshape.p[is.inf.j.mix] <- tmp7b[is.inf.j.mix]
tmp8 <- (d0A.i[, jay] - d0B.PI.mix[, jay]) / DELTA.i.mix[, jay]
dl.dpstr.mix[is.inf.j.mix] <- tmp8[is.inf.j.mix]
if (li.mix > 1) {
tmp2 <- pstr.mix * d1A.i[, jay] / DELTA.i.mix[, jay]
dl.dshape.i[is.inf.j.mix] <- tmp2[is.inf.j.mix]
}
tmp9i <- d0B.PI.mix[, jay] / DELTA.i.mix[, jay]
n.tmp <- -tmp9i[is.inf.j.mix]
p.tmp <- +tmp9i[is.inf.j.mix]
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mix, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mix ] <- p.tmp
} # jay
} # li.mix > 0
if (tmp3.TF[ 6] && ld.mix > 0) { # aka \calD_{p}
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
is.def.j.mix <- extra$skip.mix.d[, jay] # Logical vector
tmp7b <- Numer * d1B.PD.mix[, jay] / DELTA.d.mix[, jay]
dl.dshape.p[is.def.j.mix] <- tmp7b[is.def.j.mix]
tmp8 <- (d0B.PD.mix[, jay] - d0A.d[, jay]) / DELTA.d.mix[, jay]
dl.dpdip.mix[is.def.j.mix] <- tmp8[is.def.j.mix]
if (ld.mix > 1) {
tmp2 <- (-pdip.mix) * d1A.d[, jay] / DELTA.d.mix[, jay]
dl.dshape.d[is.def.j.mix] <- tmp2[is.def.j.mix]
}
tmp9d <- d0B.PD.mix[, jay] / DELTA.d.mix[, jay]
n.tmp <- -tmp9d[is.def.j.mix]
p.tmp <- +tmp9d[is.def.j.mix]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.def.j.mix, ] <- p.tmp
} # jay
} # ld.mix > 0
new.ansd <- matrix(0, n, M) # Same dimension as eta
tmp3.TF <- !is.na(rowSums(extra$indeta))
if (lall.len) { # An MLM fitted
all6.dldp <- cbind(if (tmp3.TF[ 2]) dl.dpobs.mix else NULL,
if (tmp3.TF[ 4]) dl.dpstr.mix else NULL,
if (tmp3.TF[ 6]) dl.dpdip.mix else NULL,
if (tmp3.TF[ 8]) dl.dpobs.mlm else NULL,
if (tmp3.TF[ 9]) dl.dpstr.mlm else NULL,
if (tmp3.TF[10]) dl.dpdip.mlm else NULL)
rSs.tmp <- rowSums(allprobs[, -ncol(allprobs), drop = FALSE] *
all6.dldp)
new.ansd[, -ind.shape.z] <- allprobs[, -ncol(allprobs)] *
(all6.dldp - rSs.tmp)
} # lall.len
dshape.p.deta <- dtheta.deta(shape.p, .lshape.p , .eshape.p )
if (tmp3.TF[ 3])
dshape.a.deta <- dtheta.deta(shape.a, .lshape.a , .eshape.a )
if (tmp3.TF[ 5])
dshape.i.deta <- dtheta.deta(shape.i, .lshape.i , .eshape.i )
if (tmp3.TF[ 7])
dshape.d.deta <- dtheta.deta(shape.d, .lshape.d , .eshape.d )
new.ansd[, 1] <- dl.dshape.p * dshape.p.deta
if (tmp3.TF[ 3])
new.ansd[, extra$indeta[3, 1]] <- dl.dshape.a * dshape.a.deta
if (tmp3.TF[ 5])
new.ansd[, extra$indeta[5, 1]] <- dl.dshape.i * dshape.i.deta
if (tmp3.TF[ 7])
new.ansd[, extra$indeta[7, 1]] <- dl.dshape.d * dshape.d.deta
onecoln.indeta <- extra$indeta[1:7, ] # One coln params only
onecoln.indeta <- na.omit(onecoln.indeta) # Only those present
allcnames <- c(rownames(onecoln.indeta),
as.character(c(a.mlm, i.mlm, d.mlm)))
colnames(new.ansd) <- allcnames
c(w) * new.ansd
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.truncate = truncate, .max.support = max.support ))),
weight = eval(substitute(expression({ # gaitdlog
wz <- matrix(0, n, M * (M + 1) / 2) # The complete size
mean.true.p <- A8.p * shape.p / (1 - shape.p)
cond.EY.p <- c(mean.true.p - Bits[["SumT1.p"]] -
Bits[["SumI1.mlm.p"]] - Bits[["SumI1.mix.p"]] -
Bits[["SumD1.mlm.p"]] - Bits[["SumD1.mix.p"]] - # 20211109
Bits[["SumA1.mlm.p"]] - Bits[["SumA1.mix.p"]]) / c(
Denom0.p -
Bits[["SumD0.mix.p"]] - Bits[["SumD0.mlm.p"]] - # 20211109
Bits[["SumI0.mix.p"]] - Bits[["SumI0.mlm.p"]])
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom0.p)
if (min(probns) < 0 || 1 < max(probns))
stop("variable 'probns' for P(nonspecial) is out of range")
zero0n <- numeric(n)
ned2l.dpobs.mix.shape.p <- zero0n # mB overwritten below [4279]
ned2l.dpobs.mix.shape.a <- zero0n # Fini; (2, 3) element
ned2l.dpobs.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpobs.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.p <- zero0n # Optional (1, 4) element
ned2l.dpstr.mix.shape.a <- zero0n # Final; nothing to do
ned2l.dpstr.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpdip.mix.shape.p <- zero0n # Optional (1, 6) element
posn.pobs.mix <- as.vector(extra$indeta[ 2, 'launch'])
posn.shape.a <- as.vector(extra$indeta[ 3, 'launch'])
posn.pstr.mix <- as.vector(extra$indeta[ 4, 'launch'])
posn.shape.i <- as.vector(extra$indeta[ 5, 'launch'])
posn.pdip.mix <- as.vector(extra$indeta[ 6, 'launch'])
posn.shape.d <- as.vector(extra$indeta[ 7, 'launch'])
posn.pobs.mlm <- as.vector(extra$indeta[ 8, 'launch'])
posn.pstr.mlm <- as.vector(extra$indeta[ 9, 'launch'])
posn.pdip.mlm <- as.vector(extra$indeta[10, 'launch'])
ned2l.dpdip.mix2 <- # Elt (6, 6)
ned2l.dpstr.mix2 <- # Elt (4, 4). Unchanged by deflation.
ned2l.dpobs.mlm.pstr.mix <- # Elts (4, >=8). (((09)))
ned2l.dpobs.mix.pstr.mix <- +probns / Numer^2 # ccc Elt (2, 4)
if (all(c(la.mix, li.mlm) > 0)) # (((08)))
ned2l.dpobs.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(li.mix, li.mlm) > 0)) # (((10)))
ned2l.dpstr.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(ld.mix, ld.mlm) > 0)) # (((21)))
ned2l.dpdip.mix.pdip.mlm <- matrix( probns / Numer^2, n, ld.mlm)
ned2l.dpobs.mlm.pdip.mix <- # Elts (6, >=8). (((19)))
ned2l.dpstr.mix.pdip.mix <- # Elt (4, 6)
ned2l.dpobs.mix.pdip.mix <- -probns / Numer^2 # ccc Elt (2, 6)
if (all(c(la.mix, ld.mlm) > 0)) # (((17)))
ned2l.dpobs.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(li.mix, ld.mlm) > 0)) # (((18)))
ned2l.dpstr.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(ld.mix, li.mlm) > 0)) # (((20)))
ned2l.dpdip.mix.pstr.mlm <- matrix(-probns / Numer^2, n, li.mlm)
ned2l.dshape.p2 <- probns * (cond.EY.p / shape.p^2 + # ccc
A8.p * (1 - A8.p) / (1 - shape.p)^2 +
Denom2.p / Denom0.p - (Denom1.p / Denom0.p)^2) +
(if (tmp3.TF[ 4] && li.mix) Numer *
rowSums(Numer * (d1B.PI.mix^2) / DELTA.i.mix - d2B.PI.mix)
else 0) +
(if (tmp3.TF[ 9] && li.mlm) Numer *
rowSums(Numer * (d1B.PI.mlm^2) / DELTA.i.mlm - d2B.PI.mlm)
else 0) +
(if (tmp3.TF[ 6] && ld.mix) Numer *
rowSums(Numer * (d1B.PD.mix^2) / DELTA.d.mix - d2B.PD.mix)
else 0) +
(if (tmp3.TF[10] && ld.mlm) Numer * # nnn.
rowSums(Numer * (d1B.PD.mlm^2) / DELTA.d.mlm - d2B.PD.mlm)
else 0)
wz[, iam(1, 1, M)] <- ned2l.dshape.p2 * dshape.p.deta^2
ned2l.dpobs.mix2 <- 1 / pobs.mix + probns / Numer^2
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpobs.mix2 <- # More just below, ccc
ned2l.dpobs.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpobs.mix2 <- # ccc.
ned2l.dpobs.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (tmp3.TF[ 6] && ld.mix > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (tmp3.TF[ 2] && la.mix > 0)
wz[, iam(2, 2, M)] <- ned2l.dpobs.mix2 # Link done later
if (tmp3.TF[ 3] && la.mix > 1) {
ned2l.dshape.a2 <- pobs.mix * (
rowSums((DA.mix.1mat.a^2) / DA.mix.0mat.a) / Denom0.a -
(Denom1.a / Denom0.a)^2) # ccc.
wz[, iam(3, 3, M)] <- ned2l.dshape.a2 * dshape.a.deta^2
}
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 +
rowSums((d0A.i - d0B.PI.mix)^2 / DELTA.i.mix)
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mix * (1 + Numer * (d0A.i - d0B.PI.mix) / DELTA.i.mix))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (all(tmp3.TF[c(2, 4)]))
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix +
rowSums(-d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
} # (tmp3.TF[ 4] && li.mix > 0)
if (all(tmp3.TF[c(2, 4, 9)])) { # was la.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 4, 6)])) { # == ld.mix > 0 & DELTA.d.mix
ned2l.dpobs.mix.pstr.mix <- # nnn
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(2, 4, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pstr.mix <- # nnn.
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pstr.mix))
wz[, iam(posn.pobs.mix, posn.pstr.mix, M)] <-
ned2l.dpobs.mix.pstr.mix # Link done later
if (all(tmp3.TF[c(2, 6)]))
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix +
rowSums( d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
if (all(tmp3.TF[c(2, 6, 9)])) { # == li.mlm > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 6, 4)])) { # == li.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (all(tmp3.TF[c(2, 6, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pdip.mix <- # nnn.
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pdip.mix))
wz[, iam(posn.pobs.mix, posn.pdip.mix, M)] <-
ned2l.dpobs.mix.pdip.mix # Link done later
if (tmp3.TF[ 5] && li.mix > 1) { # \calI_{p}, includes \theta_i.
ned2l.dshape.p.shape.i <- pstr.mix * Numer *
rowSums(d1A.i * d1B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(1, posn.shape.i, M)] <- ned2l.dshape.p.shape.i *
dshape.p.deta * dshape.i.deta # All links done here
ned2l.dshape.i2 <- pstr.mix *
rowSums(pstr.mix * (d1A.i^2) / DELTA.i.mix - d2A.i) # ccc.
wz[, iam(posn.shape.i, posn.shape.i, M)] <-
ned2l.dshape.i2 * dshape.i.deta^2
if (tmp3.TF[ 2]) { # tmp3.TF[ 4] is TRUE, given tmp3.TF[ 5]
ned2l.dpobs.mix.shape.i <-
rowSums(-pstr.mix * d1A.i * d0B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(posn.pobs.mix, posn.shape.i, M)] <-
ned2l.dpobs.mix.shape.i # * dshape.i.deta done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.i <- rowSums( # ccc.
d1A.i * (pstr.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix - 1))
wz[, iam(posn.pstr.mix, posn.shape.i, M)] <-
ned2l.dpstr.mix.shape.i # * dshape.i.deta done later
}
if (all(tmp3.TF[c(5, 6)])) {
ned2l.dpdip.mix.shape.i <- rowSums(
(-pstr.mix) * d0B.PI.mix * d1A.i / DELTA.i.mix)
wz[, iam(posn.pdip.mix, posn.shape.i, M)] <-
ned2l.dpdip.mix.shape.i # link done later
}
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.i <- rowSums(
-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix) # ccc.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.i, M)] <-
ned2l.dpobs.mlm.shape.i # * dshape.i.deta done later
}
} # (tmp3.TF[ 5] && li.mix > 1)
if (tmp3.TF[ 6] && ld.mix > 0) { # \calD_{p}, maybe w. \theta_d
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mix * (1 + Numer * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 +
rowSums((d0A.d - d0B.PD.mix)^2 / DELTA.d.mix)
} # (tmp3.TF[ 6] && ld.mix > 0)
if (tmp3.TF[ 7] && ld.mix > 1) { # \calD_{p}, includes \theta_d
ned2l.dshape.p.shape.d <- (-pdip.mix) * Numer *
rowSums(d1A.d * d1B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(1, posn.shape.d, M)] <- ned2l.dshape.p.shape.d *
dshape.p.deta * dshape.d.deta # All links done here
if (tmp3.TF[ 2]) { # tmp3.TF[ 6] is TRUE, given tmp3.TF[ 7]
ned2l.dpobs.mix.shape.d <-
rowSums(pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(posn.pobs.mix, posn.shape.d, M)] <-
ned2l.dpobs.mix.shape.d # link done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.d <- rowSums(
pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix)
wz[, iam(posn.pstr.mix, posn.shape.d, M)] <-
ned2l.dpstr.mix.shape.d # * dshape.i.deta done later
}
ned2l.dpdip.mix.shape.d <- rowSums(
d1A.d * (1 + pdip.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
wz[, iam(posn.pdip.mix, posn.shape.d, M)] <-
ned2l.dpdip.mix.shape.d # * dshape.d.deta done later
ned2l.dshape.d2 <- pdip.mix *
rowSums(pdip.mix * (d1A.d^2) / DELTA.d.mix + d2A.d) # nnn.
wz[, iam(posn.shape.d, posn.shape.d, M)] <-
ned2l.dshape.d2 * dshape.d.deta^2
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.d <- rowSums(
pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix) # nnn.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.d, M)] <-
ned2l.dpobs.mlm.shape.d # * dshape.d.deta done later
}
} # (tmp3.TF[ 7] && ld.mix > 1)
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # ccc
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ned2l.dpstr.mix.shape.p <- # ccc.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
} # tmp3.TF[ 9] && li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # nnn.
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpstr.mix.shape.p <- # nnn.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
} # tmp3.TF[10] && ld.mlm > 0
if (!is.na(posn.pobs.mix)) # Optional (1, 2) element:
wz[, iam(1, posn.pobs.mix, M)] <-
ned2l.dpobs.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix)) # Optional (1, 4) element
wz[, iam(1, posn.pstr.mix, M)] <-
ned2l.dpstr.mix.shape.p # One link done later
if (!is.na(posn.pdip.mix)) # Optional (1, 6) element
wz[, iam(1, posn.pdip.mix, M)] <-
ned2l.dpdip.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix) &&
!is.na(posn.pdip.mix)) # Optional (4, 6) element
wz[, iam(posn.pstr.mix, posn.pdip.mix, M)] <-
ned2l.dpstr.mix.pdip.mix # Links done later zz1
if (!is.na(posn.pstr.mix)) # Optional (4, 4) element
wz[, iam(posn.pstr.mix, # Link done later
posn.pstr.mix, M)] <- ned2l.dpstr.mix2
if (!is.na(posn.pdip.mix)) # Optional (6, 6) element
wz[, iam(posn.pdip.mix, # Link done later
posn.pdip.mix, M)] <- ned2l.dpdip.mix2
if (tmp3.TF[ 8] && la.mlm) { # \calA_{np}, includes \omega_s
ofset <- posn.pobs.mlm - 1 # 7 for GAITD combo
for (uuu in seq(la.mlm)) { # Diagonal elts only
wz[, iam(ofset + uuu,
ofset + uuu, M)] <- 1 / pobs.mlm[, uuu]
} # uuu
tmp8a <- probns / Numer^2
if (tmp3.TF[ 4] && li.mix)
tmp8a <- tmp8a + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
tmp8a <- tmp8a + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
tmp8a <- tmp8a + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
tmp8a <- tmp8a + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # All elts
for (vvv in uuu:la.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
wz[, iam(ofset + uuu, ofset + vvv, M)] + tmp8a # All elts
} # la.mlm
if (tmp3.TF[ 8] && la.mlm) {
init0.i.val <- init0.d.val <- 0
if (tmp3.TF[ 9] && li.mlm) init0.i.val <-
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[10] && ld.mlm) init0.d.val <-
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpobs.mlm.shape.p <- init0.i.val + init0.d.val # Vector
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums( # nnn
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ofset <- posn.pobs.mlm - 1 # 5 for combo
for (vvv in seq(la.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpobs.mlm.shape.p
} # la.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s
init0.val <- probns / Numer^2
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (ld.mix) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (ld.mlm) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
ned2l.dpstr.mlm2 <-
matrix(init0.val, n, li.mlm * (li.mlm + 1) / 2)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss])^2 / DELTA.i.mlm[, sss]
if (li.mlm > 1) {
for (uuu in seq(li.mlm - 1))
for (vvv in (uuu + 1):li.mlm)
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
} # if (li.mlm > 1)
ofset <- posn.pstr.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in uuu:li.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s
init0.val <- probns / Numer^2
if (ld.mix)
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (li.mlm)
init0.val <- init0.val + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
ned2l.dpdip.mlm2 <-
matrix(init0.val, n, ld.mlm * (ld.mlm + 1) / 2)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu))^2 / DELTA.d.mlm[, sss]
if (ld.mlm > 1) {
for (uuu in seq(ld.mlm - 1))
for (vvv in (uuu + 1):ld.mlm)
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu)) *
(d0B.PD.mlm[, sss] - (sss == vvv)) / DELTA.d.mlm[, sss]
} # if (ld.mlm > 1)
ofset <- posn.pdip.mlm - 1
for (uuu in seq(ld.mlm))
for (vvv in uuu:ld.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)]
} # ld.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpstr.mlm.theta.p <- matrix(0, n, li.mlm)
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.theta.p[, vvv] <-
ned2l.dpstr.mlm.theta.p[, vvv] +
d1B.PI.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PI.mlm[, sss]) / (
DELTA.i.mlm[, sss]))
if (li.mix && tmp3.TF[ 4])
ned2l.dpstr.mlm.theta.p <-
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpstr.mlm.theta.p <- # nnn
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (ld.mlm && tmp3.TF[10])
ned2l.dpstr.mlm.theta.p <- # nnn.
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ofset <- posn.pstr.mlm - 1
for (vvv in seq(li.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpstr.mlm.theta.p[, vvv]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpdip.mlm.theta.p <- matrix(0, n, ld.mlm)
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm.theta.p[, vvv] <-
ned2l.dpdip.mlm.theta.p[, vvv] - # Minus
d1B.PD.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PD.mlm[, sss]) / (
DELTA.d.mlm[, sss]))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (li.mix && tmp3.TF[ 4])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (li.mlm && tmp3.TF[ 9])
ned2l.dpdip.mlm.theta.p <- # nnn.
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ofset <- posn.pdip.mlm - 1
for (vvv in seq(ld.mlm)) # nnn.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpdip.mlm.theta.p[, vvv]
} # ld.mlm > 0
if (li.mlm && li.mix > 1) {
ned2l.dpstr.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.i, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.i # ccc.
} # li.mlm && li.mix > 1
if (ld.mlm && ld.mix > 1) {
ned2l.dpdip.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.d, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.d # nnn.
} # ld.mlm && ld.mix > 1
if (ld.mlm && li.mix > 1) {
ned2l.dpdip.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.i, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.i # nnn.
} # ld.mlm && li.mix > 1
if (li.mlm && ld.mix > 1) {
ned2l.dpstr.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.d, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.d # nnn.
} # li.mlm && ld.mix > 1
if (all(c(la.mlm, li.mlm) > 0)) {
ned2l.dpobs.mlm.pstr.mlm <-
array(probns / Numer^2, c(n, la.mlm, li.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] - d0B.PI.mlm[, sss] *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pstr.mlm <-
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pstr <- posn.pstr.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pstr + vvv, M)] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(li.mlm, ld.mlm) > 0)) {
ned2l.dpstr.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, li.mlm, ld.mlm))
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PI.mlm[, sss] *
((sss == uuu) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpstr.mlm.pdip.mlm <-
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpstr.mlm.pdip.mlm <- # nnn.
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pstr <- posn.pstr.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pstr + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv]
} # all(c(li.mlm, ld.mlm) > 0)
if (all(c(la.mlm, ld.mlm) > 0)) {
ned2l.dpobs.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, la.mlm, ld.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(la.mix, la.mlm) > 0)) {
ned2l.dpobs.mix.pobs.mlm <- probns / Numer^2 # Initialize
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 7]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pobs.mix, posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pobs.mlm # Link done later
}
if (all(c(la.mix, li.mlm) > 0)) { # all(tmp3.TF[c(2, 9)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pstr.mlm <-
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pstr.mlm <- # nnn
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mix.pstr.mlm <- # nnn; + is correct, not -
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mix.pstr.mlm[, uuu] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm)) # ccc.
wz[, iam(posn.pobs.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] # Link done later
} # all(c(la.mix, li.mlm) > 0)
if (all(c(la.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(2, 10)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mix.pdip.mlm[, uuu] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
for (uuu in seq(ld.mlm)) # nnn.
wz[, iam(posn.pobs.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] # Link done later
} # all(c(la.mix, ld.mlm) > 0)
if (all(c(li.mix, la.mlm) > 0)) { # all(tmp3.TF[c(4, 8)])
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pstr.mix <-
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ned2l.dpobs.mlm.pstr.mix <- # tmp3.TF[ 4] && li.mix
ned2l.dpobs.mlm.pstr.mix -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pstr.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pstr.mix # Link done later
} # all(c(li.mix, la.mlm) > 0
if (all(c(ld.mix, la.mlm) > 0)) { # all(tmp3.TF[c(6, 8)])
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
ned2l.dpobs.mlm.pdip.mix <- # all(tmp3.TF[c(6, 8)])
ned2l.dpobs.mlm.pdip.mix +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
for (uuu in seq(la.mlm)) # nnn.
wz[, iam(posn.pdip.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pdip.mix # Link done later
} # all(c(ld.mix, la.mlm) > 0
if (all(c(li.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpstr.mix.pstr.mlm[, uuu] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
ned2l.dpstr.mix.pstr.mlm <-
ned2l.dpstr.mix.pstr.mlm -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] # Link done later
} # all(c(li.mix, li.mlm) > 0
if (all(c(ld.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(6, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[ 9]
for (sss in seq(ld.mlm))
ned2l.dpdip.mix.pdip.mlm[, uuu] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] -
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm -
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] # Link done later
} # all(c(ld.mix, ld.mlm) > 0
if (all(c(ld.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpdip.mix.pstr.mlm[, uuu] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] # Link done later
} # all(c(ld.mix, li.mlm) > 0
if (all(c(li.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(4, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[10]
for (sss in seq(ld.mlm))
ned2l.dpstr.mix.pdip.mlm[, uuu] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (li.mix) # tmp3.TF[ 4]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm +
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpstr.mix.pdip.mlm <- # nnn.
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] # Link done later
} # all(c(li.mix, ld.mlm) > 0)
if (lall.len) {
wz.6 <- matrix(0, n, M * (M + 1) / 2) # Or == 0 * wz
ind.rc <- setdiff(1:M, ind.shape.z) # Contiguous rows and
lind.rc <- length(ind.rc) # cols of the DAMLM
# Copy in the thetas values: the looping is overkill.
for (uuu in ind.shape.z)
for (sss in seq(M))
wz.6[, iam(uuu, sss, M)] <- wz[, iam(uuu, sss, M)]
speed.up <- intercept.only && (
length(offset) == 1 || all(offset[1] == offset))
IND.mlm <- iam(NA, NA, lind.rc, both = TRUE, diag = TRUE)
n.use <- if (speed.up) 2 else n # For sandwich.mlm
if (!length(extra$ind.wz.match)) {
Imat <- matrix(NA, lind.rc, lind.rc)
for (jay in seq(lind.rc)) {
iptr <- jay
for (kay in (ind.rc[jay]):M) {
if (!any(kay %in% ind.shape.z)) {
Imat[jay, iptr] <-
which(extra$index.M$row == ind.rc[jay] &
extra$index.M$col == kay)
iptr <- iptr + 1
} # if
} # kay
} # jay
ind.wz.match <- Imat[cbind(IND.mlm$row.ind,
IND.mlm$col.ind)]
extra$ind.wz.match <- ind.wz.match # Assign it once
} # !length(extra$ind.wz.match)
filling <- if (speed.up)
wz[1:n.use, extra$ind.wz.match, drop = FALSE] else
wz[, extra$ind.wz.match, drop = FALSE]
M.mlm <- lind.rc
if (is.null(extra$iamlist)) {
extra$iamlist <- iamlist <-
iam(NA, NA, M = M.mlm, both = TRUE)
if (M.mlm > 1) { # Offdiagonal elts
extra$iamlist.nod <- iamlist.nod <-
iam(NA, NA, M.mlm, both = TRUE, diag = FALSE)
}
} # is.null(extra$iamlist)
iamlist <- extra$iamlist
iamlist.nod <- extra$iamlist.nod
MM12.mlm <- M.mlm * (M.mlm + 1) / 2
Qf3 <- rowSums(filling[, 1:M.mlm, drop = FALSE] * # Diag elts
(allprobs[1:n.use, 1:M.mlm, drop = FALSE])^2)
if (M.mlm > 1) # Offdiagonal elts
Qf3 <- Qf3 + 2 * rowSums(allprobs[1:n.use, iamlist.nod$row] *
filling[, -(1:M.mlm), drop = FALSE] * # n-vector
allprobs[1:n.use, iamlist.nod$col])
Qf3 <- matrix(Qf3, n.use, MM12.mlm)
Qf2rowsums <- matrix(0, n.use, M.mlm) # rowsums stored columnwise
for (want in seq(M.mlm)) { # Want the equivalent of rowSums(Qf2a)
iamvec <- iam(want, 1:M.mlm, M = M.mlm) # Diagonals included
Qf2rowsums[, want] <- rowSums(filling[, iamvec, drop = FALSE] *
allprobs[1:n.use, 1:M.mlm])
} # want
Qf2a <- Qf2rowsums[, iamlist$row]
Qf2b <- Qf2rowsums[, iamlist$col]
Qform <- filling - Qf2a - Qf2b + Qf3 # n x MM12.mlm
Qform <- Qform *
allprobs[1:n.use, iamlist$row, drop = FALSE] *
allprobs[1:n.use, iamlist$col, drop = FALSE]
wz.6[, extra$ind.wz.match] <- if (speed.up)
matrix(Qform[1, ], n, ncol(Qform), byrow = TRUE) else c(Qform)
dstar.deta <- cbind(dshape.p.deta,
if (tmp3.TF[ 3]) dshape.a.deta else NULL,
if (tmp3.TF[ 5]) dshape.i.deta else NULL,
if (tmp3.TF[ 7]) dshape.d.deta else NULL)
iptr <- 0
if (length(ind.shape.z))
for (uuu in ind.shape.z) { # Could delete 3 for shape.a (orthog)
iptr <- iptr + 1
for (ttt in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <- 0 # Initialize
for (sss in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <-
wz.6[, iam(uuu, ind.rc[ttt], M)] +
allprobs[, sss] * (max(0, sss == ttt) - allprobs[, ttt]) *
wz[, iam(uuu, ind.rc[sss], M)] * dstar.deta[, iptr]
} # sss
} # ttt
} # uuu
wz <- wz.6 # Completed
} # lall.len
if (lall.len) { # A MLM was fitted
mytiny <- (allprobs < sqrt(.Machine$double.eps)) |
(allprobs > 1.0 - sqrt(.Machine$double.eps))
atiny <- rowSums(mytiny) > 0
if (any(atiny)) {
ind.diags <- setdiff(1:M, ind.shape.z) # Exclude thetas
wz[atiny, ind.diags] <- .Machine$double.eps +
wz[atiny, ind.diags] * (1 + .Machine$double.eps^0.5)
}
} # lall.len
c(w) * wz
}), list( .truncate = truncate ))))
} # gaitdlog
moments.gaitdcombo.binom <-
function(size.p, prob.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
size.a = size.p,
size.i = size.p,
size.d = size.p,
prob.a = prob.p,
prob.i = prob.p,
prob.d = prob.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
if (is.infinite(max.support)) {
rmlife1 <- rmlife2 <- numeric(length(size.p)) # 0
} else {
stop("currently RML unknown for finite 'max.support'")
x.use <- max.support + 1
rmlife1 <- NA
rmlife2 <- NA
} # is.infinite(max.support)
mylist1 <-
moments.gaitdcombo.2par(
theta1.p = size.p, theta2.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta1.a = size.a, theta2.a = prob.a,
theta1.i = size.i, theta2.i = prob.i,
theta1.d = size.d, theta2.d = prob.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dgaitdbinom") # do.call() called.
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this * (munb.p - SumA1.mix.p -
SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.binom
dgaitdbinom <-
function(x, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, # vector
pobs.mlm = 0, # matrix
pstr.mix = 0, # vector
pstr.mlm = 0, # matrix
pdip.mix = 0, # vector
pdip.mlm = 0, # matrix
byrow.aid = FALSE, # Applies to 'pobs.mlm' & 'pstr.mlm'
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
log = FALSE,
...) { # ... is for max.support (ignored)
max.support <- Inf
log.arg <- log; rm(log)
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0 # Lower support
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(dbinom(x, size.p, prob.p, log = log.arg))
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(x),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a),
length(size.i), length(size.d),
length(prob.p), length(prob.a),
length(prob.i), length(prob.d))
if (length(x) < LLL) x <- rep_len(x, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
sumt <- 0 # Initialization to 0 important
if (ltrunc)
for (tval in truncate)
sumt <- sumt + dbinom(tval, size.p, prob.p)
vecTF.t <- is.finite(x) & ((x %in% truncate) | (max.support < x))
cdf.max.s <- pbinom(max.support, size.p, prob.p) # Usually 1
denom.t <- cdf.max.s - sumt # No sumt on RHS
pmf0 <- ifelse(vecTF.t, 0, dbinom(x, size.p, prob.p) / denom.t)
sum.a <- suma <- 0 # numeric(LLL)
vecTF.a <- rep_len(FALSE, LLL)
if (la.mlm) {
pobs.mlm <- matrix(pobs.mlm, LLL, la.mlm, byrow = byrow.aid)
sum.a <- .rowSums(pobs.mlm, LLL, la.mlm)
if (any(1 < sum.a, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'") # zz
for (aval in a.mlm)
suma <- suma + dbinom(aval, size.p, prob.p) # Part i
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
if (any(vecTF <- is.finite(x) & aval == x)) {
pmf0[vecTF] <- pobs.mlm[vecTF, jay]
}
vecTF.a <- vecTF.a | vecTF # Cumulative
} # jay
} # la.mlm
pmf2.a <- pmf2.i <- pmf2.d <- 0
if (la.mix) {
allx.a <- lowsup:max(a.mix)
pmf2.a <- dgaitdbinom(x, # Outer distribution---mlm type
size.a, prob.a,
truncate = setdiff(allx.a, a.mix))
for (aval in a.mix) { # Part ii added; cumulative
suma <- suma + dbinom(aval, size.p, prob.p)
vecTF <- is.finite(x) & aval == x
pmf0[vecTF] <- 0 # added; the true values are assigned below
vecTF.a <- vecTF.a | vecTF # Cumulative; added
}
} # la.mix
if (li.mix) {
allx.i <- lowsup:max(i.mix)
pmf2.i <- dgaitdbinom(x, # Outer distribution---mlm type
size.i, prob.i,
truncate = setdiff(allx.i, i.mix))
}
sum.d <- 0 # numeric(LLL)
if (ld.mlm) {
pdip.mlm <- matrix(pdip.mlm, LLL, ld.mlm, byrow = byrow.aid)
sum.d <- .rowSums(pdip.mlm, LLL, ld.mlm)
if (any(1 < sum.d, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
} # ld.mlm
if (ld.mix) {
allx.d <- lowsup:max(d.mix)
pmf2.d <- dgaitdbinom(x, size.p = size.d, prob.p = prob.d,
truncate = setdiff(allx.d, d.mix))
} # ld.mix
sum.i <- 0
if (li.mlm) {
pstr.mlm <- matrix(pstr.mlm, LLL, li.mlm, byrow = byrow.aid)
sum.i <- .rowSums(pstr.mlm, LLL, li.mlm)
if (any(1 < sum.i, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
} # li.mlm
skip <- vecTF.t | vecTF.a # Leave these values alone
tmp6 <- 1 - sum.a - sum.i - pobs.mix - pstr.mix + sum.d + pdip.mix
if (li.mlm + ld.mlm) {
if (any(tmp6[!skip] < 0, na.rm = TRUE)) {
warning("the vector of normalizing constants contains ",
"some negative values. Replacing them with NAs")
tmp6[!skip & tmp6 < 0] <- NA
}
} # li.mlm + ld.mlm
pmf0[!skip] <- (tmp6 * # added
dbinom(x, size.p, prob.p) / (cdf.max.s - suma - sumt))[!skip]
if (li.mlm) {
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
if (any(vecTF <- is.finite(x) & ival == x)) {
pmf0[vecTF] <- pmf0[vecTF] + pstr.mlm[vecTF, jay]
}
} # jay
} # li.mlm
if (ld.mlm) {
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
if (any(vecTF <- is.finite(x) & dval == x)) {
pmf0[vecTF] <- pmf0[vecTF] - pdip.mlm[vecTF, jay]
}
} # jay
} # ld.mlm
pmf0 <- pmf0 + pobs.mix * pmf2.a + pstr.mix * pmf2.i -
pdip.mix * pmf2.d
if (log.arg) log(pmf0) else pmf0
} # dgaitdbinom
pgaitdbinom <-
function(q, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0,
pobs.mlm = 0,
pstr.mix = 0,
pstr.mlm = 0,
pdip.mix = 0,
pdip.mlm = 0,
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
lower.tail = TRUE,
...) { # ... is for max.support (ignored)
max.support <- Inf
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(pbinom(q, size.p, prob.p, lower.tail = lower.tail))
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(q),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a), length(size.i),
length(size.d),
length(prob.p), length(prob.a), length(prob.i),
length(prob.d))
offset.a <- offset.i <- offset.d <-
Offset.a <- Offset.i <- Offset.d <- numeric(LLL)
if (length(q) < LLL) q <- rep_len(q, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
sumt <- 0
fudge.t <- numeric(LLL)
cdf.max.s <- pbinom(max.support, size.p, prob.p) # Usually 1
if (ltrunc) {
for (tval in truncate) {
pmf.p <- dbinom(tval, size.p, prob.p)
sumt <- sumt + pmf.p
if (any(vecTF <- is.finite(q) & tval <= q))
fudge.t[vecTF] <- fudge.t[vecTF] + pmf.p[vecTF]
}
} # ltrunc
sum.a <- suma <- 0 # numeric(LLL)
fudge.a <- numeric(LLL)
if (la.mlm) {
pobs.mlm <- matrix(pobs.mlm, LLL, la.mlm, byrow = byrow.aid)
sum.a <- .rowSums(pobs.mlm, LLL, la.mlm)
if (any(1 < sum.a, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
pmf.p <- dbinom(aval, size.p, prob.p)
suma <- suma + pmf.p # cumulative; part i
if (any(vecTF <- (is.finite(q) & aval <= q))) {
offset.a[vecTF] <- offset.a[vecTF] + pobs.mlm[vecTF, jay]
fudge.a[vecTF] <- fudge.a[vecTF] + pmf.p[vecTF] # cmtive
}
} # jay
} # la.mlm
sum.i <- 0
if (li.mlm) {
pstr.mlm <- matrix(pstr.mlm, LLL, li.mlm, byrow = byrow.aid)
sum.i <- .rowSums(pstr.mlm, LLL, li.mlm)
if (any(1 < sum.i, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
if (any(vecTF <- (is.finite(q) & ival <= q))) {
offset.i[vecTF] <- offset.i[vecTF] + pstr.mlm[vecTF, jay]
}
} # jay
} # li.mlm
use.pobs.mix <- 0
if (la.mix) {
use.pobs.mix <- matrix(0, LLL, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
pmf.a <- dbinom(aval, size.a, prob.a)
pmf.p <- dbinom(aval, size.p, prob.p)
use.pobs.mix[, jay] <- pmf.a
suma <- suma + pmf.p # cumulative; part ii
}
use.pobs.mix <- pobs.mix *
use.pobs.mix / rowSums(use.pobs.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
pmf.p <- dbinom(aval, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & aval <= q))) {
Offset.a[vecTF] <- Offset.a[vecTF] + use.pobs.mix[vecTF, jay]
fudge.a[vecTF] <- fudge.a[vecTF] + pmf.p[vecTF] # cumulative
}
} # jay
} # la.mix
use.pstr.mix <- 0
if (li.mix) {
use.pstr.mix <- matrix(0, LLL, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
use.pstr.mix[, jay] <- dbinom(ival, size.i, prob.i)
}
use.pstr.mix <- pstr.mix *
use.pstr.mix / rowSums(use.pstr.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.p <- dbinom(ival, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & ival <= q))) {
Offset.i[vecTF] <- Offset.i[vecTF] + use.pstr.mix[vecTF, jay]
}
} # jay
} # li.mix
sum.d <- 0
if (ld.mlm) {
pdip.mlm <- matrix(pdip.mlm, LLL, ld.mlm, byrow = byrow.aid)
sum.d <- .rowSums(pdip.mlm, LLL, ld.mlm)
if (any(1 < sum.d, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
if (any(vecTF <- (is.finite(q) & dval <= q))) {
offset.d[vecTF] <- offset.d[vecTF] + pdip.mlm[vecTF, jay]
}
} # jay
} # ld.mlm
use.pdip.mix <- 0
if (ld.mix) {
use.pdip.mix <- matrix(0, LLL, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
use.pdip.mix[, jay] <- dbinom(dval, size.d, prob.d)
}
use.pdip.mix <- pdip.mix *
use.pdip.mix / rowSums(use.pdip.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.p <- dbinom(dval, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & dval <= q))) {
Offset.d[vecTF] <- Offset.d[vecTF] + use.pdip.mix[vecTF, jay]
}
} # jay
} # ld.mix
numer1 <- 1 - sum.i - sum.a - pstr.mix - pobs.mix +
sum.d + pdip.mix
denom1 <- cdf.max.s - sumt - suma
ans <- numer1 * (pbinom(q, size.p, prob.p) - fudge.t -
fudge.a) / denom1 +
offset.a + offset.i - offset.d +
Offset.a + Offset.i - Offset.d
ans[max.support <= q] <- 1
ans[ans < 0] <- 0 # Occasional roundoff error
if (lower.tail) ans else 1 - ans
} # pgaitdbinom
qgaitdbinom <-
function(p, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0,
pobs.mlm = 0,
pstr.mix = 0,
pstr.mlm = 0,
pdip.mix = 0,
pdip.mlm = 0,
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
...) { # ... is for max.support (ignored)
max.support <- NULL # Different from Inf
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(qbinom(p, size.p, prob.p )) # lower.tail, log.p = FALSE
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(p),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a),
length(size.i), length(size.d),
length(prob.p), length(prob.a),
length(prob.i), length(prob.d))
if (length(p) < LLL) p <- rep_len(p, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
pobs.mlm <- matrix(pobs.mlm, LLL, max(la.mlm, 1),
byrow = byrow.aid)
pstr.mlm <- matrix(pstr.mlm, LLL, max(li.mlm, 1),
byrow = byrow.aid)
pdip.mlm <- matrix(pdip.mlm, LLL, max(ld.mlm, 1),
byrow = byrow.aid)
min.support <- lowsup # Usual case; same as lowsup
min.support.use <- if (ltrunc)
min(setdiff(min.support:(ltrunc+5), truncate)) else
min.support
ans <- p + size.p + size.a + size.i + size.d +
prob.p + prob.a + prob.i + prob.d
bad0.p <- !is.finite(size.p) | size.p <= 0 |
!is.finite(prob.p) | prob.p <= 0 | 1 <= prob.p
bad0.a <- !is.finite(size.a) | size.a <= 0 |
!is.finite(prob.a) | prob.a <= 0 | 1 <= prob.a
bad0.i <- !is.finite(size.i) | size.i <= 0 |
!is.finite(prob.i) | prob.i <= 0 | 1 <= prob.i
bad0.d <- !is.finite(size.d) | size.d <= 0 |
!is.finite(prob.d) | prob.d <= 0 | 1 <= prob.d
bad0 <- bad0.p | bad0.a | bad0.i | bad0.d
bad <- bad0 | !is.finite(p) | p <= 0 | 1 <= p
Lo <- rep_len(min.support.use - 0.5, LLL)
approx.ans <- Lo # True at lhs
Hi <- rep_len(max.support + 0.5, LLL) # Need finite RHS endpoint
dont.iterate <- bad
done <- dont.iterate |
p <= pgaitdbinom(Hi, size.p, prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix, pobs.mix = pobs.mix,
pdip.mix = pdip.mix,
pstr.mlm = pstr.mlm, pobs.mlm = pobs.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i,
size.d = size.d,
prob.a = prob.a, prob.i = prob.i,
prob.d = prob.d,
byrow.aid = FALSE)
iter <- 0
max.iter <- round(log2(.Machine$double.xmax)) - 3
while (!all(done) && iter < max.iter) {
Lo[!done] <- Hi[!done]
Hi[!done] <- 2 * Hi[!done] + 10.5 # Bug fixed
Hi <- pmin(max.support + 0.5, Hi) # 20190924
done[!done] <-
(p[!done] <= pgaitdbinom(Hi[!done],
size.p[!done], prob.p[!done],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix[!done],
pstr.mix = pstr.mix[!done],
pdip.mix = pdip.mix[!done],
pobs.mlm = pobs.mlm[!done, , drop = FALSE],
pstr.mlm = pstr.mlm[!done, , drop = FALSE],
pdip.mlm = pdip.mlm[!done, , drop = FALSE],
size.a = size.a[!done],
size.i = size.i[!done],
size.d = size.d[!done],
prob.a = prob.a[!done],
prob.i = prob.i[!done],
prob.d = prob.d[!done],
byrow.aid = FALSE))
iter <- iter + 1
}
foo <- function(q, size.p, prob.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, pstr.mix = 0, pdip.mix = 0,
pobs.mlm = 0, pstr.mlm = 0, pdip.mlm = 0,
size.a = size.p, size.i = size.p,
size.d = size.p,
prob.a = prob.p, prob.i = prob.p,
prob.d = prob.p,
byrow.aid = FALSE, p)
pgaitdbinom(q, size.p = size.p, prob.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, prob.a = prob.a,
size.i = size.i, prob.i = prob.i,
size.d = size.d, prob.d = prob.d,
byrow.aid = FALSE) - p
lhs <- dont.iterate |
p <= dgaitdbinom(min.support.use,
size.p = size.p, prob.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, prob.a = prob.a,
size.i = size.i, prob.i = prob.i,
size.d = size.d, prob.d = prob.d,
byrow.aid = FALSE)
if (any(!lhs)) {
approx.ans[!lhs] <-
bisection.basic(foo, Lo[!lhs], Hi[!lhs], tol = 1/16,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE,
p = p[!lhs])
faa <- floor(approx.ans[!lhs])
tmp <-
ifelse(pgaitdbinom(faa,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE) < p[!lhs] &
p[!lhs] <= pgaitdbinom(faa + 1,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE),
faa + 1, faa)
ans[!lhs] <- tmp
} # any(!lhs)
if (ltrunc)
while (any(vecTF <- !bad & ans %in% truncate))
ans[vecTF] <- 1 + ans[vecTF]
vecTF <- !bad0 & !is.na(p) &
p <= dgaitdbinom(min.support.use, size.p, prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i,
size.d = size.d,
prob.a = prob.a, prob.i = prob.i,
prob.d = prob.d,
byrow.aid = FALSE)
ans[vecTF] <- min.support.use
ans[!bad0 & !is.na(p) & p == 0] <- min.support.use
ans[!bad0 & !is.na(p) & p == 1] <- max.support # Inf
ans[!bad0 & !is.na(p) & p < 0] <- NaN
ans[!bad0 & !is.na(p) & p > 1] <- NaN
ans[ bad0] <- NaN
ans
} # qgaitdbinom
rgaitdbinom <-
function(n, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, # vector
pobs.mlm = 0, # matrix
pstr.mix = 0, # vector
pstr.mlm = 0, # matrix
pdip.mix = 0, # vector
pdip.mlm = 0, # matrix
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
...) { # ... is for max.support (ignored)
qgaitdbinom(runif(n), size.p, prob.p,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix,
pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i, size.d = size.d,
prob.a = prob.a, prob.i = prob.i, prob.d = prob.d,
byrow.aid = byrow.aid)
} # rgaitdbinom
gaitd.errorcheck <-
function(a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL,
max.support = Inf,
min.support = 0, nparams = 1) {
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (!is.numeric(max.support) || is.na(max.support) ||
length(max.support) != 1 || max.support < min.support ||
round(max.support) != max.support ||
(length(truncate) && (
min(truncate, na.rm = TRUE) < min.support ||
max.support <= max(truncate, na.rm = TRUE))))
stop("bad input for argument 'max.support' and/or ",
"'truncate'")
allargs <- c(a.mix, a.mlm, i.mix, i.mlm, d.mix, d.mlm)
allargs <- c(allargs, truncate) # No NA, NaN, -Inf or Inf allowed
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm)
if (!is.Numeric(allargs, integer.valued = TRUE) ||
any(allargs < min.support) ||
any(max.support < allargs))
stop("bad input for arguments 'a.mix', 'a.mlm', ",
"'i.mix', 'i.mlm', 'd.mix' and/or 'd.mlm'")
if (length(unique(allargs)) < la.mix + la.mlm + li.mix + li.mlm +
ld.mix + ld.mlm + ltrunc)
stop("duplicate values found in arguments 'a.mix', ",
"'a.mlm', 'i.mix', 'i.mlm', 'd.mix', 'd.mlm'",
" and 'truncate'")
if (nparams == 2) {
if(la.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(a.mix) == ", la.mix,
" points")
if (li.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(i.mix) == ", li.mix,
" points")
if (ld.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(d.mix) == ", ld.mix,
" points")
}
} # gaitd.errorcheck
moments.gaitdcombo.1par <-
function(theta.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, # Vector
pobs.mlm = 0, # Matrix
pstr.mix = 0,
pstr.mlm = 0, # Ditto
pdip.mix = 0,
pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
theta.a = theta.p, theta.i = theta.p, theta.d = theta.p,
moments2 = FALSE, # Use this for variances.
rmlife1 = 0, rmlife2 = 0,
dfun = "dpois") {
NOS <- 1
nnn <- length(theta.p)
pfun <- dfun
substring(pfun, 1) <- "p" # Replace the "d" by a "p"
cdf.max.s <- do.call(pfun, list(max.support, theta.p))
LALT.MIX <- length(a.mix)
LALT.MLM <- length(a.mlm)
LINF.MIX <- length(i.mix)
LINF.MLM <- length(i.mlm)
LDEF.MIX <- length(d.mix)
LDEF.MLM <- length(d.mlm)
LTRUNCAT <- length(truncate)
if (LALT.MLM == 0) {
if (!all(pobs.mlm == 0))
warning("ignoring argument 'pobs.mlm'")
pobs.mlm <- 0
}
if (LINF.MLM == 0) {
if (!all(pstr.mlm == 0))
warning("ignoring argument 'pstr.mlm'")
pstr.mlm <- 0
}
if (LDEF.MLM == 0) {
if (!all(pdip.mlm == 0))
warning("ignoring argument 'pdip.mlm'")
pdip.mlm <- 0
}
if (LALT.MIX == 0) {
if (!all(pobs.mix == 0))
warning("ignoring argument 'pobs.mix'")
pobs.mix <- 0
}
if (LINF.MIX == 0) {
if (!all(pstr.mix == 0))
warning("ignoring argument 'pstr.mix'")
pstr.mix <- 0
}
if (LDEF.MIX == 0) {
if (!all(pdip.mix == 0))
warning("ignoring argument 'pdip.mix'")
pdip.mix <- 0
}
SumT0.p <- matrix(0, nnn, NOS) # Does not include upper RHS tail
SumT1.p <- matrix(rmlife1, nnn, NOS) # Includes RHS tail
SumT2.p <- matrix(rmlife2, nnn, NOS) # Includes RHS tail
if (LTRUNCAT)
for (tval in truncate) {
pmf.p <- do.call(dfun, list(tval, theta.p))
SumT0.p <- SumT0.p + pmf.p # Need tval<=max.support
SumT1.p <- SumT1.p + pmf.p * tval
if (moments2)
SumT2.p <- SumT2.p + pmf.p * tval^2
}
use.pobs.mix <- use.pobs.mlm <- # So rowSums() works below.
use.pstr.mix <- use.pstr.mlm <-
use.pdip.mix <- use.pdip.mlm <- matrix(0, nnn, 1)
aprd1.mix <- aprd1.mlm <- # aprd1.m?? is an innerprod
aprd2.mix <- aprd2.mlm <- 0 # aprd2.m?? is an innerprod
SumA0.mix.p <- SumA0.mlm.p <-
SumA0.mix.a <- SumA0.mlm.a <-
SumA1.mix.p <- SumA1.mlm.p <-
SumA1.mix.a <- SumA1.mlm.a <-
SumA1.mlm.x <-
SumA2.mix.p <- SumA2.mlm.p <-
SumA2.mix.a <- SumA2.mlm.a <-
SumA2.mlm.x <- matrix(0, nnn, NOS)
if (LALT.MIX)
use.pobs.mix <- matrix(pobs.mix, nnn, 1)
if (LINF.MIX)
use.pstr.mix <- matrix(pstr.mix, nnn, 1)
if (LDEF.MIX)
use.pdip.mix <- matrix(pdip.mix, nnn, 1)
if (LALT.MLM)
use.pobs.mlm <- matrix(pobs.mlm, nnn, LALT.MLM,
byrow = byrow.aid)
if (LINF.MLM)
use.pstr.mlm <- matrix(pstr.mlm, nnn, LINF.MLM,
byrow = byrow.aid)
if (LDEF.MLM)
use.pdip.mlm <- matrix(pdip.mlm, nnn, LDEF.MLM,
byrow = byrow.aid)
if (LALT.MIX) {
for (jay in seq_len(LALT.MIX)) {
aval <- a.mix[jay]
pmf.p <- do.call(dfun, list(aval, theta.p))
pmf.a <- do.call(dfun, list(aval, theta.a))
SumA0.mix.p <- SumA0.mix.p + pmf.p
SumA0.mix.a <- SumA0.mix.a + pmf.a
SumA1.mix.p <- SumA1.mix.p + pmf.p * aval
SumA1.mix.a <- SumA1.mix.a + pmf.a * aval
if (moments2) {
SumA2.mix.p <- SumA2.mix.p + pmf.p * aval^2
SumA2.mix.a <- SumA2.mix.a + pmf.a * aval^2
}
} # for jay
aprd1.mix <- use.pobs.mix * SumA1.mix.a / SumA0.mix.a
if (moments2)
aprd2.mix <- use.pobs.mix * SumA2.mix.a / SumA0.mix.a
} # LALT.MIX
if (LALT.MLM) {
for (jay in seq_len(LALT.MLM)) {
aval <- a.mlm[jay]
pmf.x <- use.pobs.mlm[, jay]
pmf.p <- do.call(dfun, list(aval, theta.p))
pmf.a <- do.call(dfun, list(aval, theta.a))
SumA0.mlm.p <- SumA0.mlm.p + pmf.p
SumA0.mlm.a <- SumA0.mlm.a + pmf.a
SumA1.mlm.p <- SumA1.mlm.p + pmf.p * aval
SumA1.mlm.a <- SumA1.mlm.a + pmf.a * aval
SumA1.mlm.x <- SumA1.mlm.x + pmf.x * aval
if (moments2) {
SumA2.mlm.p <- SumA2.mlm.p + pmf.p * aval^2
SumA2.mlm.a <- SumA2.mlm.a + pmf.a * aval^2
SumA2.mlm.x <- SumA2.mlm.x + pmf.x * aval^2
}
} # for jay
aprd1.mlm <- SumA1.mlm.x
if (moments2)
aprd2.mlm <- SumA2.mlm.x
} # LALT.MLM
iprd1.mix <- iprd1.mlm <- # iprd1.m?? is an innerprod
iprd2.mix <- iprd2.mlm <- 0 # iprd2.m?? is an innerprod
SumI0.mix.p <- SumI0.mlm.p <-
SumI0.mix.i <- SumI0.mlm.i <-
SumI1.mix.p <- SumI1.mlm.p <-
SumI1.mix.i <- SumI1.mlm.i <-
SumI1.mlm.x <-
SumI2.mix.p <- SumI2.mlm.p <-
SumI2.mix.i <- SumI2.mlm.i <-
SumI2.mlm.x <- matrix(0, nnn, NOS)
dprd1.mix <- dprd1.mlm <-
dprd2.mix <- dprd2.mlm <- 0
SumD0.mix.p <- SumD0.mlm.p <-
SumD0.mix.d <- SumD0.mlm.d <-
SumD1.mix.p <- SumD1.mlm.p <-
SumD1.mix.d <- SumD1.mlm.d <-
SumD1.mlm.x <-
SumD2.mix.p <- SumD2.mlm.p <-
SumD2.mix.d <- SumD2.mlm.d <-
SumD2.mlm.x <- matrix(0, nnn, NOS)
if (LINF.MIX) {
for (jay in seq_len(LINF.MIX)) {
ival <- i.mix[jay]
pmf.p <- do.call(dfun, list(ival, theta.p))
pmf.i <- do.call(dfun, list(ival, theta.i))
SumI0.mix.p <- SumI0.mix.p + pmf.p
SumI0.mix.i <- SumI0.mix.i + pmf.i
SumI1.mix.p <- SumI1.mix.p + pmf.p * ival
SumI1.mix.i <- SumI1.mix.i + pmf.i * ival
if (moments2) {
SumI2.mix.p <- SumI2.mix.p + pmf.p * ival^2
SumI2.mix.i <- SumI2.mix.i + pmf.i * ival^2
}
} # for jay
iprd1.mix <- use.pstr.mix * SumI1.mix.i / SumI0.mix.i
if (moments2)
iprd2.mix <- use.pstr.mix * SumI2.mix.i / SumI0.mix.i
} # LINF.MIX
if (LINF.MLM) {
for (jay in seq_len(LINF.MLM)) {
ival <- i.mlm[jay]
pmf.x <- use.pstr.mlm[, jay]
pmf.p <- do.call(dfun, list(ival, theta.p))
pmf.i <- do.call(dfun, list(ival, theta.i))
SumI0.mlm.p <- SumI0.mlm.p + pmf.p
SumI0.mlm.i <- SumI0.mlm.i + pmf.i
SumI1.mlm.p <- SumI1.mlm.p + pmf.p * ival
SumI1.mlm.i <- SumI1.mlm.i + pmf.i * ival
SumI1.mlm.x <- SumI1.mlm.x + pmf.x * ival
if (moments2) {
SumI2.mlm.p <- SumI2.mlm.p + pmf.p * ival^2
SumI2.mlm.i <- SumI2.mlm.i + pmf.i * ival^2
SumI2.mlm.x <- SumI2.mlm.x + pmf.x * ival^2
}
} # for jay
iprd1.mlm <- SumI1.mlm.x
if (moments2)
iprd2.mlm <- SumI2.mlm.x
} # LINF.MLM
if (LDEF.MIX) {
for (jay in seq_len(LDEF.MIX)) {
dval <- d.mix[jay]
pmf.p <- do.call(dfun, list(dval, theta.p))
pmf.d <- do.call(dfun, list(dval, theta.d))
SumD0.mix.p <- SumD0.mix.p + pmf.p
SumD0.mix.d <- SumD0.mix.d + pmf.d
SumD1.mix.p <- SumD1.mix.p + pmf.p * dval
SumD1.mix.d <- SumD1.mix.d + pmf.d * dval
if (moments2) {
SumD2.mix.p <- SumD2.mix.p + pmf.p * dval^2
SumD2.mix.d <- SumD2.mix.d + pmf.d * dval^2
}
} # for jay
dprd1.mix <- use.pdip.mix * SumD1.mix.d / SumD0.mix.d
if (moments2)
dprd2.mix <- use.pdip.mix * SumD2.mix.d / SumD0.mix.d
} # LDEF.MIX
if (LDEF.MLM) {
for (jay in seq_len(LDEF.MLM)) {
dval <- d.mlm[jay]
pmf.x <- use.pdip.mlm[, jay]
pmf.p <- do.call(dfun, list(dval, theta.p))
pmf.d <- do.call(dfun, list(dval, theta.d))
SumD0.mlm.p <- SumD0.mlm.p + pmf.p
SumD0.mlm.d <- SumD0.mlm.d + pmf.d
SumD1.mlm.p <- SumD1.mlm.p + pmf.p * dval
SumD1.mlm.d <- SumD1.mlm.d + pmf.d * dval
SumD1.mlm.x <- SumD1.mlm.x + pmf.x * dval
if (moments2) {
SumD2.mlm.p <- SumD2.mlm.p + pmf.p * dval^2
SumD2.mlm.d <- SumD2.mlm.d + pmf.d * dval^2
SumD2.mlm.x <- SumD2.mlm.x + pmf.x * dval^2
}
} # for jay
dprd1.mlm <- SumD1.mlm.x
if (moments2)
dprd2.mlm <- SumD2.mlm.x
} # LDEF.MLM
use.this <- 1 - rowSums(use.pobs.mlm) - rowSums(use.pstr.mlm) +
rowSums(use.pdip.mlm) -
use.pobs.mix - use.pstr.mix + use.pdip.mix
ans <- list('cdf.max.s' = cdf.max.s,
'SumT0.p' = SumT0.p,
'SumT1.p' = SumT1.p,
'SumA0.mix.a' = SumA0.mix.a,
'SumA0.mix.p' = SumA0.mix.p,
'SumA1.mix.a' = SumA1.mix.a,
'SumA1.mix.p' = SumA1.mix.p,
'SumA0.mlm.a' = SumA0.mlm.a,
'SumA0.mlm.p' = SumA0.mlm.p,
'SumA1.mlm.a' = SumA1.mlm.a,
'SumA1.mlm.p' = SumA1.mlm.p,
'SumI0.mix.i' = SumI0.mix.i,
'SumI0.mix.p' = SumI0.mix.p,
'SumI1.mix.i' = SumI1.mix.i,
'SumI1.mix.p' = SumI1.mix.p,
'SumI0.mlm.i' = SumI0.mlm.i,
'SumI0.mlm.p' = SumI0.mlm.p,
'SumI1.mlm.i' = SumI1.mlm.i,
'SumI1.mlm.p' = SumI1.mlm.p,
'SumD0.mix.d' = SumD0.mix.d,
'SumD0.mix.p' = SumD0.mix.p,
'SumD1.mix.d' = SumD1.mix.d,
'SumD1.mix.p' = SumD1.mix.p,
'SumD0.mlm.d' = SumD0.mlm.d,
'SumD0.mlm.p' = SumD0.mlm.p,
'SumD1.mlm.d' = SumD1.mlm.d,
'SumD1.mlm.p' = SumD1.mlm.p,
'aprd1.mix' = aprd1.mix,
'aprd1.mlm' = aprd1.mlm,
'iprd1.mix' = iprd1.mix,
'iprd1.mlm' = iprd1.mlm,
'dprd1.mix' = dprd1.mix, #
'dprd1.mlm' = dprd1.mlm, #
'use.this' = use.this)
if (moments2) { # Add more info
ans <- c(ans,
list( # 'rmlife2' = rmlife2, # May be scalar
'aprd2.mix' = aprd2.mix,
'aprd2.mlm' = aprd2.mlm,
'iprd2.mix' = iprd2.mix,
'iprd2.mlm' = iprd2.mlm,
'dprd2.mix' = dprd2.mix, #
'dprd2.mlm' = dprd2.mlm, #
'SumT2.p' = SumT2.p,
'SumA2.mix.p' = SumA2.mix.p,
'SumA2.mix.a' = SumA2.mix.a,
'SumI2.mix.p' = SumI2.mix.p,
'SumI2.mix.i' = SumI2.mix.i,
'SumD2.mix.p' = SumD2.mix.p, #
'SumD2.mix.d' = SumD2.mix.d, #
'SumA2.mlm.p' = SumA2.mlm.p,
'SumA2.mlm.a' = SumA2.mlm.a,
'SumI2.mlm.p' = SumI2.mlm.p,
'SumI2.mlm.i' = SumI2.mlm.i,
'SumD2.mlm.p' = SumD2.mlm.p, #
'SumD2.mlm.d' = SumD2.mlm.d)) #
}
ans
} # moments.gaitdcombo.1par
moments.gaitdcombo.pois <-
function(lambda.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, # Vector
pobs.mlm = 0, # Matrix
pstr.mix = 0,
pstr.mlm = 0, # Ditto
pdip.mix = 0,
pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
lambda.a = lambda.p, lambda.i = lambda.p,
lambda.d = lambda.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
rmlife1 <- ppois(max.support - 1, lambda.p, lower.tail = FALSE) *
lambda.p
rmlife2 <- ppois(max.support - 2, lambda.p, lower.tail = FALSE) *
lambda.p^2 + rmlife1
mylist1 <- moments.gaitdcombo.1par(theta.p = lambda.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = lambda.a, theta.i = lambda.i,
theta.d = lambda.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dpois")
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this * (lambda.p - SumA1.mix.p -
SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.pois
moments.gaitdcombo.log <-
function(shape.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
shape.a = shape.p, shape.i = shape.p, shape.d = shape.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
A8.p <- -1 / log1p(-shape.p)
rmlife1 <- A8.p * (shape.p^(max.support + 1)) / (1 - shape.p)
rmlife2 <- A8.p * ((shape.p^(max.support + 1)) *
(max.support + 1 / (1 - shape.p))
/ (1 - shape.p))
mylist1 <- moments.gaitdcombo.1par(theta.p = shape.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = shape.a, theta.i = shape.i, theta.d = shape.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dlog")
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this *
(-shape.p / (log1p(-shape.p) * (1 - shape.p)) -
SumA1.mix.p - SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.log
moments.gaitdcombo.zeta <-
function(shape.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
shape.a = shape.p, shape.i = shape.p, shape.d = shape.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
rmlife1 <- if (is.finite(max.support)) zeta(shape.p) * (1 -
pzeta(max.support, shape.p - 1)) / zeta(shape.p + 1) else
numeric(length(shape.p))
rmlife1[shape.p <= 1] <- NA # NA or Inf, not sure
rmlife2 <- if (is.finite(max.support)) zeta(shape.p - 1) * (1 -
pzeta(max.support, shape.p - 2)) / zeta(shape.p + 1) else
numeric(length(shape.p))
rmlife2[shape.p <= 2] <- NA # NA or Inf, not sure
mylist1 <- moments.gaitdcombo.1par(theta.p = shape.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = shape.a, theta.i = shape.i, theta.d = shape.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dzeta")
themean <-
with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this *
(ifelse(shape.p > 1, zeta(shape.p) / zeta(shape.p + 1), NA) -
SumA1.mix.p - SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.zeta
specialsvglm <-
function(object, ...) {
infos <- object@family@infos()
ans <- list(a.mix = infos$a.mix,
a.mlm = infos$a.mlm,
i.mix = infos$i.mix,
i.mlm = infos$i.mlm,
d.mix = infos$d.mix,
d.mlm = infos$d.mlm,
truncate = infos$truncate)
if (is.numeric(tmp7e <- infos$max.support))
ans <- c(ans, max.support = tmp7e)
ans
} # specialsvglm
if (!isGeneric("specials"))
setGeneric("specials", function(object, ...)
standardGeneric("specials"),
package = "VGAM")
setMethod("specials", "vglm",
function(object, ...)
specialsvglm(object, ...))
if (!isGeneric("altered"))
setGeneric("altered", function(object, ...)
standardGeneric("altered"),
package = "VGAM")
setMethod("altered", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
c(tmp$a.mix, tmp$a.mlm)})
if (!isGeneric("inflated"))
setGeneric("inflated", function(object, ...)
standardGeneric("inflated"),
package = "VGAM")
setMethod("inflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
c(tmp$i.mix, tmp$i.mlm)})
if (!isGeneric("truncated"))
setGeneric("truncated", function(object, ...)
standardGeneric("truncated"),
package = "VGAM")
setMethod("truncated", "vglm",
function(object, ...) {
ans <- specialsvglm(object, ...)
if (any(names(ans) == "max.support"))
ans[c("truncate", "max.support")] else
ans[["truncate"]]
})
setGeneric("is.altered", function(object, ...)
standardGeneric("is.altered"),
package = "VGAM")
setMethod("is.altered", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$a.mix, tmp$a.mlm)))})
setGeneric("is.inflated", function(object, ...)
standardGeneric("is.inflated"),
package = "VGAM")
setMethod("is.inflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$i.mix, tmp$i.mlm)))})
setGeneric("is.deflated", function(object, ...)
standardGeneric("is.deflated"),
package = "VGAM")
setMethod("is.deflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$d.mix, tmp$d.mlm)))})
setGeneric("is.truncated", function(object, ...)
standardGeneric("is.truncated"),
package = "VGAM")
setMethod("is.truncated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(tmp$truncated)) ||
(length(tmp$max.support) > 0 && is.finite(tmp$max.support))
})
y.gaitcombo.check <-
function(y, truncate = NULL,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
max.support = Inf, min.support = 0) {
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
n <- length(y)
css.mix.a <- css.mix.i <- css.mix.d <-
css.mlm.a <- css.mlm.i <- css.mlm.d <- NULL
skip.mix.a <- skip.mix.i <- skip.mix.d <- # Default
skip.mlm.a <- skip.mlm.i <- skip.mlm.d <- NULL
if (length(truncate) && any(y %in% truncate))
stop("some response values == values in argument 'truncate'")
if (max.support < max(y))
stop("some response values are greater than the ",
"'max.support' argument")
y0.mix.a <- y0.mlm.a <-
y0.mix.i <- y0.mlm.i <-
y0.mix.d <- y0.mlm.d <- NULL
if (la.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix))
y0.mix.a[, jay] <- as.numeric(y == a.mix[jay])
skip.mix.a <- matrix(as.logical(y0.mix.a), n, la.mix) # dim lost
if (any((css.mix.a <- colSums(skip.mix.a)) == 0))
stop("some 'a.mix' argument values have no response values: ",
paste(a.mix[css.mix.a == 0], collapse = ", "))
} # la.mix
if (la.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.a <- matrix(0, n, la.mlm)
for (jay in seq(la.mlm))
y0.mlm.a[, jay] <- as.numeric(y == a.mlm[jay])
skip.mlm.a <- matrix(as.logical(y0.mlm.a), n, la.mlm) # dim lost
if (any((css.mlm.a <- colSums(skip.mlm.a)) == 0))
stop("some 'a.mlm' argument values have no response values: ",
paste(a.mlm[css.mlm.a == 0], collapse = ", "))
} # la.mlm
if (li.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix))
y0.mix.i[, jay] <- as.numeric(y == i.mix[jay])
skip.mix.i <- matrix(as.logical(y0.mix.i), n, li.mix) # dim lost
if (any((css.mix.i <- colSums(skip.mix.i)) == 0))
stop("some 'i.mix' argument values have no response values: ",
paste(i.mix[css.mix.i == 0], collapse = ", "))
} # li.mix
if (li.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm))
y0.mlm.i[, jay] <- as.numeric(y == i.mlm[jay])
skip.mlm.i <- matrix(as.logical(y0.mlm.i), n, li.mlm) # dim lost
if (any((css.mlm.i <- colSums(skip.mlm.i)) == 0))
stop("some 'i.mlm' argument values have no response values: ",
paste(i.mlm[css.mlm.i == 0], collapse = ", "))
} # li.mlm
if (ld.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix))
y0.mix.d[, jay] <- as.numeric(y == d.mix[jay])
skip.mix.d <- matrix(as.logical(y0.mix.d), n, ld.mix) # dim lost
if (any((css.mix.d <- colSums(skip.mix.d)) == 0))
stop("some 'd.mix' argument values have no response values: ",
paste(d.mix[css.mix.d == 0], collapse = ", "))
} # ld.mix
if (ld.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm))
y0.mlm.d[, jay] <- as.numeric(y == d.mlm[jay])
skip.mlm.d <- matrix(as.logical(y0.mlm.d), n, ld.mlm) # dim lost
if (any((css.mlm.d <- colSums(skip.mlm.d)) == 0))
stop("some 'd.mlm' argument values have no response values: ",
paste(d.mlm[css.mlm.d == 0], collapse = ", "))
} # ld.mlm
list(css.mix.a = css.mix.a, skip.mix.a = skip.mix.a,
css.mix.i = css.mix.i, skip.mix.i = skip.mix.i,
css.mix.d = css.mix.d, skip.mix.d = skip.mix.d,
css.mlm.a = css.mlm.a, skip.mlm.a = skip.mlm.a,
css.mlm.i = css.mlm.i, skip.mlm.i = skip.mlm.i,
css.mlm.d = css.mlm.d, skip.mlm.d = skip.mlm.d,
y0.mix.a = y0.mix.a, y0.mlm.a = y0.mlm.a,
y0.mix.i = y0.mix.i, y0.mlm.i = y0.mlm.i,
y0.mix.d = y0.mix.d, y0.mlm.d = y0.mlm.d)
} # y.gaitcombo.check
Try the VGAM package in your browser
Any scripts or data that you put into this service are public.
VGAM documentation built on Sept. 18, 2024, 9:09 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions y.gaitcombo.check specialsvglm moments.gaitdcombo.zeta moments.gaitdcombo.log moments.gaitdcombo.pois moments.gaitdcombo.1par gaitd.errorcheck rgaitdbinom qgaitdbinom pgaitdbinom dgaitdbinom moments.gaitdcombo.binom gaitdlog gaitdzeta Trunc meangaitd get.indices.gaitd pd.damlm.old pd.damlm
Documented in dgaitdbinom gaitd.errorcheck gaitdlog gaitdzeta get.indices.gaitd meangaitd moments.gaitdcombo.1par moments.gaitdcombo.binom moments.gaitdcombo.log moments.gaitdcombo.pois moments.gaitdcombo.zeta pd.damlm pgaitdbinom qgaitdbinom rgaitdbinom specialsvglm Trunc y.gaitcombo.check
# These functions are
# Copyright (C) 1998-2024 T.W. Yee, University of Auckland.
# All rights reserved.
amazon.col <- "#3b7a57"
avocado.col <- "#568203"
indigo.col <- "#6D5ACF" # Indigo http://hexcolor16.com/6d5acf
iris.col <- "#5a4fcf"
turquoise.col <- "#30d5c8" # for truncation
dirt.col <- "#9b7653"
deer.col <- "#ba8759"
desire.col <- "#ea3c53" # A dark red colour
peach.col <- "#ffe5b4"
azure.col <- "#007fff" # Unneeded
artichoke.col <- "#8f9779"
asparagus.col <- "#87a96b"
pd.damlm <-
function(ind.num, ind.den,
prob.num, prob.den = prob.num,
is.dipped = c(FALSE, FALSE)) {
if (is.dipped[1] && !is.dipped[2])
is.dipped <- c(FALSE, FALSE)
if (!any(is.dipped)) {
return(prob.num * ((ind.num == ind.den) - prob.den))
}
if (!is.dipped[1] && is.dipped[2]) # Negated.
return( prob.num * prob.den)
if (all(is.dipped)) # Negated.
return(-((ind.num == ind.den) - prob.num) * prob.den)
stop("am confused... should never reach here")
} # pd.damlm
if (FALSE)
pd.damlm.old <-
function(num.a = NULL, num.d = NULL,
den.a = NULL, den.d = NULL,
prob.num, prob.den = prob.num,
Denom = NA, eta.d.max = 0) {
if (length(num.d) && length(den.d)) {
if (num.a == num.d)
return((1 - prob.num) * (prob.num - exp(eta.d.max) / Denom)) else
return(-prob.num * (prob.den - exp(eta.d.max) / Denom))
}
if (length(num.a) && length(den.d)) {
return(-prob.num * (prob.den - exp(eta.d.max) / Denom))
}
if (length(num.a) && length(den.a)) {
if (num.a == num.d) return(prob.num * (1 - prob.num))
}
return(-prob.num * prob.den)
} # pd.damlm.old
get.indices.gaitd <- function(ind, indeta) {
ind.b <- indeta[ind, 'launch']
ind.b <- c(na.omit(ind.b))
ind.e <- indeta[ind, 'finish']
ind.e <- c(na.omit(ind.e))
ind.z <- NULL
if (length(ind.e) > 0)
for (jay in seq(length(ind.e)))
ind.z <- c(ind.z, seq(from = ind.b[jay], to = ind.e[jay]))
ind.z
} # get.ind.gaitd
meangaitd <-
function(theta.p,
fam = c("pois", "log", "zeta"), # "genpois0",
a.mix = NULL, i.mix = NULL, d.mix = NULL, #
a.mlm = NULL, i.mlm = NULL, d.mlm = NULL, #
truncate = NULL, max.support = Inf,
pobs.mix = 0, # scalar
pobs.mlm = 0, # vector of length a.mlm
pstr.mix = 0, # scalar
pstr.mlm = 0, # vector of length a.mlm
pdip.mix = 0, # scalar
pdip.mlm = 0, # vector of length d.mlm
byrow.aid = FALSE, # Applies to 'pobs.mlm' & 'pstr.mlm'
theta.a = theta.p, # scalar, else a 2-vector
theta.i = theta.p, # scalar, else a 2-vector
theta.d = theta.p, # scalar, else a 2-vector
...
) { # ... ignored currently.
fam.choices <- c("pois", "log", "zeta")
fam <- match.arg(fam[1], fam.choices)[1]
baseparams.argnames <-
switch(fam, "pois" = "lambda", "log" = "shape", "zeta" = "shape")
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
MM <- switch(fam, "pois" = 1, "log" = 1, "zeta" = 1)
if (MM != 1 && MM != 2)
stop("can only handle 1 or 2 parameters")
if (!length(a.mix)) pobs.mix <- 0 # Make sure for all
if (!length(a.mlm)) pobs.mlm <- 0
if (!length(i.mix)) pstr.mix <- 0
if (!length(i.mlm)) pstr.mlm <- 0
if (!length(d.mix)) pdip.mix <- 0
if (!length(d.mlm)) pdip.mlm <- 0
if (length(pobs.mix) != 1) stop("bad input for argument 'pobs.mix'")
if (length(pstr.mix) != 1) stop("bad input for argument 'pstr.mix'")
if (length(pdip.mix) != 1) stop("bad input for argument 'pdip.mix'")
if (length(a.mlm) && length(pobs.mlm) > length(a.mlm))
warning("bad input for argument 'pobs.mlm'?")
if (length(i.mlm) && length(pstr.mlm) > length(i.mlm))
warning("bad input for argument 'pstr.mlm'?")
if (length(d.mlm) && length(pdip.mlm) > length(d.mlm))
warning("bad input for argument 'pdip.mlm'?")
if (length(a.mlm))
pobs.mlm <- matrix(pobs.mlm, 1, # length(xx),
length(a.mlm), byrow = byrow.aid)
if (length(i.mlm))
pstr.mlm <- matrix(pstr.mlm, 1, # length(xx),
length(i.mlm), byrow = byrow.aid)
if (length(d.mlm))
pdip.mlm <- matrix(pdip.mlm, 1, # length(xx),
length(d.mlm), byrow = byrow.aid)
alist <- list( # x = xx, # theta.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid)
alist[[paste0(baseparams.argnames[1], ".p")]] <- theta.p[1]
alist[[paste0(baseparams.argnames[1], ".a")]] <- theta.a[1]
alist[[paste0(baseparams.argnames[1], ".i")]] <- theta.i[1]
alist[[paste0(baseparams.argnames[1], ".d")]] <- theta.d[1]
if (MM == 2) {
alist[[paste0(baseparams.argnames[2], ".p")]] <- theta.p[2]
alist[[paste0(baseparams.argnames[2], ".a")]] <- theta.a[2]
alist[[paste0(baseparams.argnames[2], ".i")]] <- theta.i[2]
alist[[paste0(baseparams.argnames[2], ".d")]] <- theta.d[2]
}
mlist <- alist
mlist$type.fitted <- "All"
mlist$moments2 <- TRUE
mom.fun <- paste0("moments.gaitdcombo.", fam)
Bits <- do.call(mom.fun, mlist)
if (length(c(a.mix, a.mlm, i.mix, i.mlm, d.mix, d.mlm))) {
Denom.p <- as.vector(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p <= 0))
stop("0s found in the denominator (variable 'Denom.p')")
Numer <- as.vector(1 -
(if (length(a.mix)) pobs.mix else 0) -
(if (length(i.mix)) pstr.mix else 0) +
(if (length(d.mix)) pdip.mix else 0) -
(if (length(a.mlm)) rowSums(rbind(pobs.mlm)) else 0) -
(if (length(i.mlm)) rowSums(rbind(pstr.mlm)) else 0) +
(if (length(d.mlm)) rowSums(rbind(pdip.mlm)) else 0))
if (!all(is.finite(Numer)))
warning("variable 'Numer' contains non-finite values")
if (min(Numer, na.rm = TRUE) < 0)
warning("variable 'Numer' has negative values")
} # Inflation or deflation
c(Bits$mean)
} # meangaitd
Trunc <- function(Range, mux = 2, location = 0, omits = TRUE) {
if (!is.finite(mux) || length(mux) != 1 || round(mux) != mux ||
mux < 1)
stop("argument 'mux' must be a positive integer")
if (any(!is.finite(Range)) || length(Range) < 2 ||
!all(round(Range) == Range))
stop("bad input in argument 'Range'")
if (length(Range) > 2)
Range <- range(Range, na.rm = TRUE) # For vglm() and na.omit()
Min <- Range[1]
Max <- Range[2]
allx <- location + (mux * Min):(mux * Max)
multiples <- location + mux * (Min:Max)
if (omits) {
ans <- setdiff(allx, multiples)
if (length(ans)) ans else NULL
} else {
multiples
}
} # Trunc
gaitdzeta <-
function(a.mix = NULL, i.mix = NULL,
d.mix = NULL,
a.mlm = NULL, i.mlm = NULL, # Unstructured probs are
d.mlm = NULL, # contiguous
truncate = NULL, max.support = Inf,
zero = c("pobs", "pstr", "pdip"), # Pruned, handles all 6
eq.ap = TRUE, eq.ip = TRUE, eq.dp = TRUE,
parallel.a = FALSE, parallel.i = FALSE,
parallel.d = FALSE,
lshape.p = "loglink",
lshape.a = lshape.p, # "logitlink", 20201117
lshape.i = lshape.p, # "logitlink", 20201117
lshape.d = lshape.p, # "logitlink", 20211011
type.fitted = c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"),
gshape.p = -expm1(-ppoints(7)),
gpstr.mix = ppoints(7) / 3, # ppoints(9) / 2,
gpstr.mlm = ppoints(7) / (3 + length(i.mlm)),
imethod = 1,
mux.init = c(0.75, 0.5, 0.75), # Order is A, I, D.
ishape.p = NULL, ishape.a = ishape.p,
ishape.i = ishape.p, ishape.d = ishape.p,
ipobs.mix = NULL, ipstr.mix = NULL, # 0.25,
ipdip.mix = NULL, # 0.01, # Easy but inflexible 0.01
ipobs.mlm = NULL, ipstr.mlm = NULL, # 0.25,
ipdip.mlm = NULL, # 0.01, # NULL, Easy but inflexible
byrow.aid = FALSE,
ishrinkage = 0.95,
probs.y = 0.35) {
mux.init <- rep_len(mux.init, 3)
if (length(a.mix) == 0) a.mix <- NULL
if (length(i.mix) == 0) i.mix <- NULL
if (length(d.mix) == 0) d.mix <- NULL
if (length(a.mlm) == 0) a.mlm <- NULL
if (length(i.mlm) == 0) i.mlm <- NULL
if (length(d.mlm) == 0) d.mlm <- NULL
if (length(truncate) == 0) truncate <- NULL
lowsup <- 1
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support,
min.support = lowsup)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltruncat <- length(truncate <- sort(truncate))
ltrunc.use <- ltruncat > 0 || !is.infinite(max.support)
if (is.finite(max.support))
stop("argument 'max.support' must be 'Inf'.")
lshape.p <- as.list(substitute(lshape.p))
eshape.p <- link2list(lshape.p)
lshape.p <- attr(eshape.p, "function.name")
lshape.p.save <- lshape.p
lpobs.mix <- "multilogitlink" # \omega_p
epobs.mix <- list() # zz NULL 4 now 20200907 coz 'multilogitlink'
eshape.a <- link2list(lshape.a)
lshape.a <- attr(eshape.a, "function.name")
lpstr.mix <- "multilogitlink" # \phi_p
epstr.mix <- list() # zz NULL 4 now 20200907 coz 'multilogitlink'
lpdip.mix <- "multilogitlink" # zz unsure 20211002
epdip.mix <- list() # zz unsure 20211002
eshape.i <- link2list(lshape.i)
lshape.i <- attr(eshape.i, "function.name")
eshape.d <- link2list(lshape.d)
lshape.d <- attr(eshape.d, "function.name")
lshape.p.save <- lshape.p
gshape.p.save <- gshape.p
if (is.vector(zero) && is.character(zero) && length(zero) == 3) {
if (li.mix + li.mlm == 0)
zero <- setdiff(zero, "pstr")
if (la.mix + la.mlm == 0)
zero <- setdiff(zero, "pobs")
if (ld.mix + ld.mlm == 0)
zero <- setdiff(zero, "pdip")
if (length(zero) == 0)
zero <- NULL # Better than character(0)
}
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
if (lall.len + ltruncat == 0 && is.infinite(max.support))
return(eval(substitute(
zetaff(lshape = .lshape.p.save ,
gshape = .gshape.p.save ,
zero = NULL),
list( .lshape.p.save = lshape.p.save,
.gshape.p.save = gshape.p.save ))))
if (!is.logical(eq.ap) || length(eq.ap) != 1)
stop("argument 'eq.ap' must be a single logical")
if (!is.logical(eq.ip) || length(eq.ip) != 1)
stop("argument 'eq.ip' must be a single logical")
if (!is.logical(parallel.a) || length(parallel.a) != 1)
stop("argument 'parallel.a' must be a single logical")
if (!is.logical(parallel.i) || length(parallel.i) != 1)
stop("argument 'parallel.i' must be a single logical")
if (!is.logical(parallel.d) || length(parallel.d) != 1)
stop("argument 'parallel.d' must be a single logical")
if (FALSE) { # Comment this out 2 allow default eq.ap=TRUE, etc.
if (la.mix <= 1 && eq.ap)
stop("<= one unstructured altered value (no 'shape.a')",
", so setting 'eq.ap = TRUE' is meaningless")
if (li.mix <= 1 && eq.ip)
stop("<= one unstructured inflated value (no 'shape.i')",
", so setting 'eq.ip = TRUE' is meaningless")
if (ld.mix <= 1 && eq.dp)
stop("<= one unstructured deflated value (no 'shape.d')",
", so setting 'eq.dp = TRUE' is meaningless")
if (la.mlm <= 1 && parallel.a) # Only \omega_1
stop("<= one altered mixture probability, 'pobs", a.mlm,
"', so setting 'parallel.a = TRUE' is meaningless")
if (li.mlm <= 1 && parallel.i) # Only \phi_1
stop("<= one inflated mixture probability, 'pstr", i.mlm,
"', so setting 'parallel.i = TRUE' is meaningless")
if (ld.mlm <= 1 && parallel.d) # Only \psi_1
stop("<= one deflated mixture probability, 'pdip", d.mlm,
"', so setting 'parallel.d = TRUE' is meaningless")
} # FALSE
type.fitted.choices <-
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s")
type.fitted <- match.arg(type.fitted[1], type.fitted.choices)[1]
tmp7a <- if (la.mlm) paste0("pobs.mlm", a.mlm) else NULL
tmp7b <- if (li.mlm) paste0("pstr.mlm", i.mlm) else NULL
tmp7c <- if (ld.mlm) paste0("pdip.mlm", d.mlm) else NULL
tmp3 <- c(shape.p = lshape.p,
pobs.mix = if (la.mix) "multilogitlink" else NULL,
shape.a = if (la.mix > 1) lshape.a else NULL,
pstr.mix = if (li.mix) "multilogitlink" else NULL,
shape.i = if (li.mix > 1) lshape.i else NULL,
pdip.mix = if (ld.mix) "multilogitlink" else NULL,
shape.d = if (ld.mix > 1) lshape.d else NULL,
if (la.mlm) rep("multilogitlink", la.mlm) else NULL,
if (li.mlm) rep("multilogitlink", li.mlm) else NULL,
if (ld.mlm) rep("multilogitlink", ld.mlm) else NULL)
Ltmp3 <- length(tmp3)
if (la.mlm + li.mlm + ld.mlm)
names(tmp3)[(Ltmp3 - la.mlm - li.mlm - ld.mlm + 1):Ltmp3] <-
c(tmp7a, tmp7b, tmp7c)
par1or2 <- 1 # 2
tmp3.TF <- c(TRUE, la.mix > 0, la.mix > 1,
li.mix > 0, li.mix > 1,
ld.mix > 0, ld.mix > 1,
la.mlm > 0, li.mlm > 0, ld.mlm > 0)
indeta.finish <- cumsum(c(par1or2, 1, par1or2,
1, par1or2,
1, par1or2,
la.mlm, li.mlm, ld.mlm,
ld.mlm + 1) * c(tmp3.TF, 1))
indeta.launch <- c(1, 1 + head(indeta.finish, -1))
indeta.launch <- head(indeta.launch, -1)
indeta.finish <- head(indeta.finish, -1)
indeta.launch[!tmp3.TF] <- NA # Not to be accessed
indeta.finish[!tmp3.TF] <- NA # Not to be accessed
indeta <- cbind(launch = indeta.launch,
finish = indeta.finish)
rownames(indeta) <- c("shape.p",
"pobs.mix", "shape.a",
"pstr.mix", "shape.i",
"pdip.mix", "shape.d",
"pobs.mlm", "pstr.mlm", "pdip.mlm")
M1 <- max(indeta, na.rm = TRUE)
predictors.names <- tmp3 # Passed into @infos and @initialize.
blurb1 <- "Z" # zz1
if (la.mlm + la.mix) blurb1 <- "Generally-altered z"
if (li.mlm + li.mix) blurb1 <- "Generally-inflated z"
if (ltrunc.use) blurb1 <- "Generally-truncated z"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && !ltrunc.use)
blurb1 <- "Generally-altered and -inflated z"
if ( (la.mlm + la.mix) && !(li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered and -truncated z"
if (!(la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-inflated and -truncated z"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered, -inflated and -truncated z"
if (ld.mlm + ld.mix) blurb1 <-
c(blurb1,
if (la.mlm + la.mix + li.mlm + li.mix) "and " else "Generally",
"-deflated ")
new("vglmff",
blurb = c(blurb1, "eta regression\n",
"(GAITD-zeta(shape.p)-",
"zeta(shape.a)-MLM-",
"zeta(shape.i)-MLM-\n",
"zeta(shape.d)-MLM generally)\n\n",
"Links: ",
namesof("shape.p", lshape.p, earg = eshape.p,
tag = FALSE),
if (la.mix > 0) c(", ", "multilogit(pobs.mix)"),
if (la.mix > 1) c(", ",
namesof("shape.a", lshape.a, eshape.a, tag = FALSE)),
if (la.mix && li.mix) ", \n ",
if (li.mix > 0) c( if (la.mix) "" else ", ",
"multilogit(pstr.mix)"),
if (li.mix > 1) c(", ",
namesof("shape.i", lshape.i, eshape.i, tag = FALSE)),
if (li.mix && ld.mix) ", \n ",
if (ld.mix > 0) c( if (li.mix) "" else ", ",
"multilogit(pdip.mix)"),
if (ld.mix > 1) c(", ",
namesof("shape.d", lshape.d, eshape.d, tag = FALSE)),
if (la.mlm) paste0(",\n",
paste0(" multilogit(", tmp7a, collapse = "),\n"),
")") else NULL,
if (li.mlm) paste0(",\n",
paste0(" multilogit(", tmp7b, collapse = "),\n"),
")") else NULL,
if (ld.mlm) paste0(",\n",
paste0(" multilogit(", tmp7c, collapse = "),\n"),
")") else NULL),
constraints = eval(substitute(expression({
M1 <- max(extra$indeta, na.rm = TRUE)
la.mix <- ( .la.mix )
li.mix <- ( .li.mix )
ld.mix <- ( .ld.mix )
la.mlm <- ( .la.mlm )
li.mlm <- ( .li.mlm )
ld.mlm <- ( .ld.mlm )
use.mat.mlm.a <- if (la.mlm) {
if ( .parallel.a ) matrix(1, la.mlm, 1) else diag(la.mlm)
} else {
NULL
}
use.mat.mlm.i <- if (li.mlm) {
if ( .parallel.i ) matrix(1, li.mlm, 1) else diag(li.mlm)
} else {
NULL
}
use.mat.mlm.d <- if (ld.mlm) {
if ( .parallel.d ) matrix(1, ld.mlm, 1) else diag(ld.mlm)
} else {
NULL
}
if (la.mlm + li.mlm + ld.mlm == 0) {
Use.mat <- use.mat.mlm <- cbind(M) # shape.p only
}
if (la.mlm + li.mlm + ld.mlm) {
nc1 <- if (length(use.mat.mlm.a)) ncol(use.mat.mlm.a) else 0
nc2 <- if (length(use.mat.mlm.i)) ncol(use.mat.mlm.i) else 0
nc3 <- if (length(use.mat.mlm.d)) ncol(use.mat.mlm.d) else 0
use.mat.mlm <- cbind(1, matrix(0, 1, nc1 + nc2 + nc3))
if (la.mlm)
use.mat.mlm <- rbind(use.mat.mlm,
cbind(matrix(0, la.mlm, 1),
use.mat.mlm.a,
if (length(use.mat.mlm.i) == 0)
NULL else matrix(0, la.mlm, nc2),
if (length(use.mat.mlm.d) == 0)
NULL else matrix(0, la.mlm, nc3)))
if (li.mlm )
use.mat.mlm <-
rbind(use.mat.mlm,
cbind(matrix(0, li.mlm, 1 + nc1),
use.mat.mlm.i,
matrix(0, li.mlm, nc3)))
if (ld.mlm)
use.mat.mlm <-
rbind(use.mat.mlm, # zz1 next line:
cbind(matrix(0, ld.mlm, 1 + nc1 + nc2),
use.mat.mlm.d))
} # la.mlm + li.mlm
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
use.mat.mix <- cm3gaitd( .eq.ap , .eq.ip , .eq.dp , npar = 1)
tmp3.subset <- tmp3.TF[-(8:10)]
use.mat.mix <- use.mat.mix[tmp3.subset, , drop = FALSE]
notall0 <- function(x) !all(x == 0)
use.mat.mix <- use.mat.mix[, apply(use.mat.mix, 2, notall0),
drop = FALSE]
if (la.mix + li.mix + ld.mix > 0)
Use.mat <- use.mat.mix
if (la.mlm + li.mlm + ld.mlm > 0) {
Use.mat <- rbind(use.mat.mix,
matrix(0, nrow(use.mat.mlm) - 1, # bottom
ncol(use.mat.mix)))
Use.mat <- cbind(Use.mat,
matrix(0, nrow(Use.mat), # RHS
ncol(use.mat.mlm) - 1))
Use.mat[row(Use.mat) > nrow(use.mat.mix) &
col(Use.mat) > ncol(use.mat.mix)] <-
use.mat.mlm[-1, -1]
} # la.mlm + li.mlm + ld.mlm > 0
if (is.null(constraints)) {
constraints <-
cm.VGAM(Use.mat, x = x, apply.int = TRUE, # FALSE
bool = .eq.ap || .eq.ip || .eq.dp ||
.parallel.a || .parallel.i || .parallel.d ,
constraints = constraints) # FALSE
} # is.null(constraints)
if (la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm)
constraints <-
cm.zero.VGAM(constraints, x = x, .zero , M = M, M1 = M1,
predictors.names = paste0(predictors.names,
names(predictors.names)))
}),
list( .zero = zero, .tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.eq.ap = eq.ap, .eq.ip = eq.ip, .eq.dp = eq.dp,
.parallel.a = parallel.a, .parallel.i = parallel.i,
.parallel.d = parallel.d,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix ))),
infos = eval(substitute(function(...) {
list(M1 = .M1 ,
Q1 = 1,
dpqrfun = "gaitdzeta",
link = .predictors.names , # ...strips... from above
link1parameter = as.logical( .lall.len <= 2), # <= 1 safer
mixture.links = any(c( .la.mlm , .li.mlm , .ld.mlm ,
.la.mix , .li.mix ,
.ld.mix ) > 1), # FALSE if NULL
a.mix = as.vector( .a.mix ), # Handles NULL
a.mlm = as.vector( .a.mlm ),
i.mix = as.vector( .i.mix ),
i.mlm = as.vector( .i.mlm ),
d.mix = as.vector( .d.mix ),
d.mlm = as.vector( .d.mlm ),
truncate = as.vector( .truncate ),
max.support = as.vector( .max.support ),
Support = c( .lowsup , Inf, 1), # a(b)c format as a,c,b.
expected = TRUE,
multipleResponses = FALSE, # zetaff can be called if TRUE
parameters.names = names( .predictors.names ),
parent.name = c("zetaff", "zeta"),
type.fitted = as.vector( .type.fitted ),
type.fitted.choices = ( .type.fitted.choices ),
baseparams.argnames = "shape",
MM1 = 1, # One parameter for 1 response (shape). Needed.
zero = .zero )
}, list( .zero = zero, .lowsup = lowsup,
.type.fitted = type.fitted,
.type.fitted.choices = type.fitted.choices,
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix,
.truncate = truncate, .max.support = max.support,
.predictors.names = predictors.names,
.M1 = M1, .lall.len = lall.len
))),
rqresslot = eval(substitute(
function(mu, y, w, eta, extra = NULL) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least 1 shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vecs
ind.shape.z <- c(na.omit(ind.shape.z)) # At least 1 value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
scrambleseed <- runif(1) # To scramble the seed
qnorm(runif(length(y),
pgaitdzeta(y - 1, shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm),
pgaitdzeta(y , shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)))
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
initialize = eval(substitute(expression({
extra$indeta <- ( .indeta ) # Avoids recomputing it
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix +
la.mlm + li.mlm + ld.mlm
truncate <- as.vector( .truncate )
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(y)
M <- NOS * M1
tmp3.TF <- ( .tmp3.TF )
temp5 <-
w.y.check(w = w, y = y,
Is.nonnegative.y = TRUE,
Is.integer.y = TRUE,
ncol.w.max = 1, # Since max.support = 9 is possible
ncol.y.max = 1,
out.wy = TRUE, colsyperw = 1, maximize = TRUE)
w <- temp5$w
y <- temp5$y
glist <- y.gaitcombo.check(y, truncate = truncate,
a.mlm = a.mlm, a.mix = a.mix,
i.mlm = i.mlm, i.mix = i.mix,
d.mlm = d.mlm, d.mix = d.mix,
max.support = .max.support ,
min.support = .min.support )
extra$skip.mix.a <- glist$skip.mix.a
extra$skip.mix.i <- glist$skip.mix.i
extra$skip.mix.d <- glist$skip.mix.d
extra$skip.mlm.a <- glist$skip.mlm.a
extra$skip.mlm.i <- glist$skip.mlm.i
extra$skip.mlm.d <- glist$skip.mlm.d
extra$NOS <- NOS <- ncoly <- ncol(y) # Number of species
extra$type.fitted <- as.vector( .type.fitted )
extra$mux.init <- as.vector( .mux.init )
extra$colnames.y <- colnames(y)
extra$M1 <- M1
extra$index.M <- iam(NA, NA, M, both = TRUE) # Used in @weight
predictors.names <- ( .predictors.names ) # Got it, named
if (!length(etastart)) {
shape.p.init <- if (length( .ishape.p )) .ishape.p else {
zetaff.Loglikfun <- function(shapeval, y, x, w, extraargs) {
sum(c(w) * dzeta(x = y, shape = shapeval, log = TRUE))
}
shape.p.grid <- ( .gshape.p )
grid.search(shape.p.grid, objfun = zetaff.Loglikfun,
y = y, w = w)
}
shape.p.init <- rep(shape.p.init, length = n)
shape.d.init <-
shape.a.init <- shape.i.init <- shape.p.init # Needed
etastart <- matrix(nrow = n, ncol = M,
theta2eta(shape.p.init, .lshape.p , earg = .eshape.p ))
mux.more.a <- extra$mux.init[1] # 0.75 Err 2 slightly smaller
init.pobs.mix <- numeric(n)
if (tmp3.TF[ 2]) { # la.mix > 0
init.pobs.mix <- if (length( .ipobs.mix )) {
rep_len( .ipobs.mix , n)
} else {
is.a.mix1 <- rowSums(extra$skip.mix.a) > 0
rep(mux.more.a * sum(w[is.a.mix1]) / sum(w), n)
}
} # la.mix > 0
if (tmp3.TF[ 3]) { # Assign coln 3; la.mix > 1
shape.a.init <- if (length( .ishape.a ))
rep_len( .ishape.a , n) else shape.p.init # A vector
etastart[, 3] <-
theta2eta(shape.a.init, .lshape.a , earg = .eshape.a )
}
init.pstr.mix <- init.pdip.mix <- numeric(n)
try.gridsearch.pstr.mix <- FALSE
if (tmp3.TF[ 4]) { # li.mix > 0
init.pstr.mix <- if (length( .ipstr.mix )) {
rep_len( .ipstr.mix , n)
} else {
try.gridsearch.pstr.mix <- TRUE
numeric(n) # Overwritten by gridsearch
}
} # li.mix > 0
if (tmp3.TF[ 5]) { # li.mix > 1
shape.i.init <- if (length( .ishape.i ))
rep_len( .ishape.i , n) else shape.p.init # A vector
etastart[, (extra$indeta[5, 'launch'])] <-
theta2eta(shape.i.init, .lshape.i , earg = .eshape.i )
} # li.mix > 1
if (tmp3.TF[ 8]) { # la.mlm
init.pobs.mlm <- if (length( .ipobs.mlm )) {
matrix( .ipobs.mlm , n, la.mlm, byrow = .byrow.aid )
} else {
mux.more.a <- extra$mux.init[1]
init.pobs.mlm <- colSums(c(w) *
extra$skip.mlm.a) / colSums(w)
init.pobs.mlm <- init.pobs.mlm * as.vector( mux.more.a )
matrix(init.pobs.mlm, n, la.mlm, byrow = TRUE)
}
} else {
init.pobs.mlm <- matrix(0, n, 1)
}
try.gridsearch.pstr.mlm <- FALSE
if (tmp3.TF[ 9]) { # li.mlm
try.gridsearch.pstr.mlm <- !(length( .ipstr.mlm ))
init.pstr.mlm <- 0 # Might be overwritten by gridsearch
if (length( .ipstr.mlm ))
init.pstr.mlm <- as.vector( .ipstr.mlm )
init.pstr.mlm <- matrix(init.pstr.mlm, n, li.mlm,
byrow = .byrow.aid )
} else {
init.pstr.mlm <- matrix(0, n, 1)
}
init.pdip.mlm <- matrix(0, n, 2) # rowSums() needs > 1 colns.
gaitdzeta.Loglikfun1.mix <-
function(pstr.mix.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mix = pstr.mix.val,
pstr.mlm = extraargs$pstr.mlm, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdzeta.Loglikfun1.mlm <-
function(pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mlm = pstr.mlm.val,
pstr.mix = extraargs$pstr.mix, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdzeta.Loglikfun2 <-
function(pstr.mix.val, pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdzeta(y, pstr.mix = pstr.mix.val,
pstr.mlm = pstr.mlm.val,
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
if (li.mix + li.mlm) {
extraargs <- list(
shape.p = shape.p.init,
shape.a = shape.a.init,
shape.i = shape.i.init,
shape.d = shape.d.init,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
pobs.mix = init.pobs.mix ,
pobs.mlm = init.pobs.mlm ,
pdip.mix = init.pdip.mix ,
pdip.mlm = init.pdip.mlm )
pre.warn <- options()$warn
options(warn = -1) # Ignore warnings during gridsearch
try.this <-
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
grid.search2( .gpstr.mix , .gpstr.mlm ,
objfun = gaitdzeta.Loglikfun2,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mix) {
extraargs$pstr.mlm <- init.pstr.mlm
grid.search ( .gpstr.mix ,
objfun = gaitdzeta.Loglikfun1.mix,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mlm) {
extraargs$pstr.mix <- init.pstr.mix
grid.search ( .gpstr.mlm ,
objfun = gaitdzeta.Loglikfun1.mlm,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
}
options(warn = pre.warn) # Restore warnings
if (any(is.na(try.this)))
warning("gridsearch returned NAs. It's going to crash.",
immediate. = TRUE)
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
init.pstr.mix <- rep_len(try.this["Value1"], n)
init.pstr.mlm <- matrix(try.this["Value2"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'",
" and/or 'gpstr.mlm = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mix) {
init.pstr.mix <- rep_len(try.this["Value"], n)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mlm) {
init.pstr.mlm <- matrix(try.this["Value"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mlm = seq(5) / 100'.")
}
} # la.mix + lnf.mix
mux.more.d <- extra$mux.init[3]
if (ld.mix) {
init.pdip.mix <- if (length( .ipdip.mix ))
rep_len( .ipdip.mix, n) else {
is.d.mix1 <- rowSums(extra$skip.mix.d) > 0
rep(mux.more.d * sum(w[is.d.mix1]) / sum(w), n)
}
} # ld.mix
if (ld.mlm) {
init.pdip.mlm <- if (length( .ipdip.mlm ))
matrix( .ipdip.mlm, n, ld.mlm, byrow = TRUE) else {
is.d.mlm1 <- rowSums(extra$skip.mlm.d) > 0
matrix(mux.more.d * (sum(w[is.d.mlm1]) / sum(w)) / ld.mlm,
n, ld.mlm)
}
} # ld.mlm
while (any((vecTF <- init.pobs.mix + init.pstr.mix + # -
init.pdip.mix +
rowSums(init.pobs.mlm) +
rowSums(init.pstr.mlm) + # -
rowSums(init.pdip.mlm) > 0.96875))) {
init.pobs.mix[vecTF] <- 0.875 * init.pobs.mix[vecTF]
init.pstr.mix[vecTF] <- 0.875 * init.pstr.mix[vecTF]
init.pdip.mix[vecTF] <- 0.875 * init.pdip.mix[vecTF]
init.pobs.mlm[vecTF, ] <- 0.875 * init.pobs.mlm[vecTF, ]
init.pstr.mlm[vecTF, ] <- 0.875 * init.pstr.mlm[vecTF, ]
init.pdip.mlm[vecTF, ] <- 0.875 * init.pdip.mlm[vecTF, ]
} # while
Numer.init1 <- 1 - rowSums(init.pobs.mlm) -
rowSums(init.pstr.mlm) - # +
rowSums(init.pdip.mlm) -
init.pobs.mix - init.pstr.mix - # +
init.pdip.mix # Differs from 'Numer'.
etastart.z <- if (lall.len == 0) NULL else {
tmp.mat <- cbind(if (tmp3.TF[ 2]) init.pobs.mix else NULL,
if (tmp3.TF[ 4]) init.pstr.mix else NULL,
if (tmp3.TF[ 6]) init.pdip.mix else NULL,
if (tmp3.TF[ 8]) init.pobs.mlm else NULL,
if (tmp3.TF[ 9]) init.pstr.mlm else NULL,
if (tmp3.TF[10]) init.pdip.mlm else NULL,
Numer.init1)
multilogitlink(tmp.mat)
} # etastart.z
if (!is.matrix(etastart.z)) etastart.z <- cbind(etastart.z)
nextone <- 1 # Might not be used actually
if (tmp3.TF[ 2]) {
etastart[, 2] <- etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 4]) { # Coln 2 or 4
etastart[, (extra$indeta[4, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 6]) { # Coln 2 or 4 or 6
etastart[, (extra$indeta[6, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 8]) {
ind8 <- (extra$indeta[8, 'launch']):(
extra$indeta[8, 'finish'])
etastart[, ind8] <- etastart.z[, nextone:(nextone +
la.mlm - 1)]
nextone <- nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (extra$indeta[9, 'launch']):(
extra$indeta[9, 'finish'])
etastart[, ind9] <- etastart.z[, nextone:(
nextone + li.mlm - 1)]
nextone <- nextone + li.mlm
}
if (tmp3.TF[10]) {
ind0 <- (extra$indeta[10, 'launch']):(
extra$indeta[10, 'finish'])
etastart[, ind0] <- etastart.z[, nextone:(nextone +
ld.mlm - 1)]
if (ncol(etastart.z) != nextone + ld.mlm - 1)
stop("miscalculation")
}
}
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.ishape.p = ishape.p,
.ishape.a = ishape.a,
.ishape.i = ishape.i,
.ishape.d = ishape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.lpdip.mix = lpdip.mix,
.epdip.mix = epdip.mix,
.ipstr.mix = ipstr.mix, .ipobs.mix = ipobs.mix,
.ipstr.mlm = ipstr.mlm, .ipobs.mlm = ipobs.mlm,
.ipdip.mix = ipdip.mix,
.ipdip.mlm = ipdip.mlm,
.byrow.aid = byrow.aid,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support,
.min.support = lowsup,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.predictors.names = predictors.names,
.mux.init = mux.init,
.gshape.p = gshape.p,
.gpstr.mix = gpstr.mix, # .gpdip.mix = gpdip.mix,
.gpstr.mlm = gpstr.mlm, # .gpdip.mlm = gpdip.mlm,
.ishrinkage = ishrinkage, .probs.y = probs.y,
.indeta = indeta,
.imethod = imethod, .type.fitted = type.fitted ))),
linkinv = eval(substitute(function(eta, extra = NULL) {
n.obs <- NROW(eta)
type.fitted <-
if (length(extra$type.fitted)) extra$type.fitted else {
warning("cannot find 'type.fitted'. Returning the 'mean'.")
"mean"
}
type.fitted <-
match.arg(type.fitted[1],
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"))[1]
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
morework <- type.fitted != "mean" # For efficiency
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(eta) / M1
Bits <- moments.gaitdcombo.zeta(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
if (morework) {
Denom.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p == 0)) {
smallval <- min(Denom.p[Denom.p > 0])
Denom.p[Denom.p == 0] <- 1e-09 # smallval
warning("0s found in variable 'Denom.p'. Trying to fix it.")
}
Numer <- c(1 - pobs.mix - pstr.mix + pdip.mix -
(if (la.mlm) rowSums(pobs.mlm) else 0) -
(if (li.mlm) rowSums(pstr.mlm) else 0) +
(if (ld.mlm) rowSums(pdip.mlm) else 0))
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom.p)
} # morework
if (!la.mlm && type.fitted %in% c("pobs.mlm")) {
warning("No altered MLM values; returning an NA")
return(NA)
}
if (!li.mlm && type.fitted %in% c("sum.mlm.i", "pstr.mlm")) {
warning("No inflated MLM values; returning an NA")
return(NA)
}
if (!ld.mlm && type.fitted %in% c("sum.mlm.d", "pdip.mlm")) {
warning("No deflated MLM values; returning an NA")
return(NA)
}
if (!la.mix && type.fitted %in% c("Pobs.mix")) {
warning("No altered mixture values; returning an NA")
return(NA)
}
if (!li.mix && type.fitted %in% c("sum.mix.i", "Pstr.mix")) {
warning("No inflated mixture values; returning an NA")
return(NA)
}
if (!ld.mix && type.fitted %in% c("sum.mix.d", "Pdip.mix")) {
warning("No deflated mixture values; returning an NA")
return(NA)
}
if (la.mix && morework) {
tmp13 <- # dpois() does not retain the matrix format
dzeta(matrix(a.mix, n.obs, la.mix, byrow = TRUE),
matrix(shape.a, n.obs, la.mix)) / (
c(Bits[["SumA0.mix.a"]]))
dim(tmp13) <- c(n.obs, la.mix)
dimnames(tmp13) <- list(rownames(eta),
as.character(a.mix))
propn.mat.a <- tmp13
} # la.mix
if (li.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dzeta(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.i, n.obs, li.mix)) / (
c(Bits[["SumI0.mix.i"]]))
dim(tmp55) <- c(n.obs, li.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(i.mix))
propn.mat.i <- tmp55 # Correct dimension
} # li.mix
if (ld.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dzeta(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.d, n.obs, ld.mix)) / (
c(Bits[["SumD0.mix.d"]]))
dim(tmp55) <- c(n.obs, ld.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(d.mix))
propn.mat.d <- tmp55 # Correct dimension
} # ld.mix
ans <- switch(type.fitted,
"mean" = Bits[["mean"]], # Unconditional mean
"shapes" = cbind(shape.p,
if (tmp3.TF[ 3]) shape.a else NULL,
if (tmp3.TF[ 5]) shape.i else NULL,
if (tmp3.TF[ 7]) shape.d else NULL),
"pobs.mlm" = pobs.mlm, # aka omegamat, n x la.mlm
"pstr.mlm" = pstr.mlm, # aka phimat, n x li.mlm
"pdip.mlm" = pdip.mlm, # aka psimat, n x ld.mlm
"pobs.mix" = pobs.mix, # n-vector
"pstr.mix" = pstr.mix, # n-vector
"pdip.mix" = pdip.mix, # n-vector
"Pobs.mix" = c(pobs.mix) * propn.mat.a, # matrix
"Pstr.mix" = c(pstr.mix) * propn.mat.i,
"Pdip.mix" = c(pdip.mix) * propn.mat.d,
"nonspecial" = probns,
"Numer" = Numer,
"Denom.p" = Denom.p,
"sum.mlm.i" = pstr.mlm + Numer *
dzeta(matrix(i.mlm, n.obs, li.mlm, byrow = TRUE),
matrix(shape.p, n.obs, li.mlm)) / Denom.p,
"sum.mlm.d" = -pdip.mlm + Numer *
dzeta(matrix(d.mlm, n.obs, ld.mlm, byrow = TRUE),
matrix(shape.p, n.obs, ld.mlm)) / Denom.p,
"sum.mix.i" = c(pstr.mix) * propn.mat.i + Numer *
dzeta(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.p, n.obs, li.mix)) / Denom.p,
"sum.mix.d" = -c(pdip.mix) * propn.mat.d + Numer *
dzeta(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.p, n.obs, ld.mix)) / Denom.p,
"ptrunc.p" = Bits[["SumT0.p"]] + 1 - Bits[["cdf.max.s"]],
"cdf.max.s" = Bits[["cdf.max.s"]]) # Pr(y <= max.support)
ynames.pobs.mlm <- as.character(a.mlm) # Works with NULLs
ynames.pstr.mlm <- as.character(i.mlm) # Works with NULLs
ynames.pdip.mlm <- as.character(d.mlm) # Works with NULLs
if (length(ans))
label.cols.y(ans, NOS = NOS, colnames.y =
switch(type.fitted,
"shapes" = c("shape.p", "shape.a", # Some colns NA
"shape.i", "shape.d")[(tmp3.TF[c(1, 3, 5, 7)])],
"Pobs.mix" = as.character(a.mix),
"sum.mix.i" = , #
"Pstr.mix" = as.character(i.mix),
"sum.mix.d" = , #
"Pdip.mix" = as.character(d.mix),
"pobs.mlm" = ynames.pobs.mlm,
"sum.mlm.i" = , #
"pstr.mlm" = ynames.pstr.mlm,
"sum.mlm.d" = , #
"pdip.mlm" = ynames.pdip.mlm,
extra$colnames.y)) else ans
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
last = eval(substitute(expression({
pred.names <- c( .predictors.names ) # Save it
link.names <- as.vector( .predictors.names )
parameter.names <- names(pred.names)
predictors.names <- NULL
for (jay in seq(M))
predictors.names <- c(predictors.names,
namesof(parameter.names[jay], link.names[jay], tag = FALSE,
earg = list())) # This isnt perfect; info is lost
misc$predictors.names <- predictors.names # Useful for coef()
misc$link <- link.names #
names(misc$link) <- parameter.names #
misc$earg <- vector("list", M1)
names(misc$earg) <- names(misc$link)
misc$earg[[1]] <- ( .eshape.p ) # First one always there
iptr <- 1
if (tmp3.TF[ 2])
misc$earg[[(iptr <- iptr + 1)]] <- list() # multilogitlink
if (tmp3.TF[ 3])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.a )
if (tmp3.TF[ 4])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 5])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.i )
if (tmp3.TF[ 6])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 7])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.d )
if (tmp3.TF[ 8]) { # la.mlm
for (ii in seq(la.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # la.mlm
if (tmp3.TF[ 9]) { # li.mlm
for (ii in seq(li.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # li.mlm
if (tmp3.TF[10]) { # ld.mlm
for (ii in seq(ld.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # ld.mlm
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.predictors.names = predictors.names,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
loglikelihood = eval(substitute(
function(mu, y, w, residuals = FALSE, eta,
extra = NULL, summation = TRUE) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
if (residuals) {
stop("loglikelihood residuals not implemented yet")
} else {
ll.elts <- c(w) *
dgaitdzeta(y, shape.p, log = TRUE, # byrow.aid = F,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)
if (summation) {
sum(ll.elts)
} else {
ll.elts
}
}
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
vfamily = c("gaitdzeta"),
validparams = eval(substitute(function(eta, y, extra = NULL) {
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
small. <- 1e-14
pobs.mix <- pstr.mix <- pdip.mix <- small. # 4 rowSums():
pobs.mlm <- pstr.mlm <- pdip.mlm <- matrix(small., NROW(eta), 1)
shape.a <- shape.i <- shape.d <- 0.5 # Needed
if (!is.matrix(eta)) eta <- as.matrix(eta)
shape.p <-
cbind(eta2theta(eta[, 1], .lshape.p , earg = .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 1] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
okay.mlm <-
all(is.finite(pobs.mlm)) && all(0 < pobs.mlm) &&
all(is.finite(pstr.mlm)) && all(0 < pstr.mlm) &&
all(is.finite(pdip.mlm)) && all(0 < pdip.mlm)
okay.mix <-
all(is.finite(shape.p)) && all(0 < shape.p) &&
all(is.finite(shape.a)) && all(0 < shape.a) &&
all(is.finite(shape.i)) && all(0 < shape.i) &&
all(is.finite(shape.d)) && all(0 < shape.d) &&
all(is.finite(pobs.mix)) && all(0 < pobs.mix) &&
all(is.finite(pstr.mix)) && all(0 < pstr.mix) &&
all(is.finite(pdip.mix)) && all(0 < pdip.mix) &&
all(pobs.mix + pstr.mix + pdip.mix +
rowSums(pobs.mlm) + rowSums(pstr.mlm) +
rowSums(pdip.mlm) < 1) # Combined
okay.mlm && okay.mix
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
simslot = eval(substitute(
function(object, nsim) {
pwts <- if (length(pwts <- object@prior.weights) > 0)
pwts else weights(object, type = "prior")
if (any(pwts != 1))
warning("ignoring prior weights")
eta <- predict(object)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
extra <- object@extra
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF )
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A AMLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
rgaitdzeta(nsim * length(shape.p), shape.p,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = .truncate , max.support = .max.support )
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
deriv = eval(substitute(expression({
tmp3.TF <- ( .tmp3.TF )
calA.p <- tmp3.TF[ 2]
calI.p <- tmp3.TF[ 4]
calD.p <- tmp3.TF[ 6]
calA.np <- tmp3.TF[ 8]
calI.np <- tmp3.TF[ 9]
calD.np <- tmp3.TF[10]
Denom1.a <- Denom1.i <- Denom1.d <-
Denom2.i <- Denom2.d <- 0 # Denom2.a is unneeded
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted.
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
minprob.baseline <- min(allprobs[, ncol(allprobs)], na.rm = TRUE)
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1) {
warning("fitted probabilities numerically 0 or 1 occurred")
} else
if (minprob.baseline < 0.10)
warning("Minimum baseline (reserve) probability close to 0")
if (control$trace)
cat("Minimum baseline (reserve) probability = ",
format(minprob.baseline, digits = 3), "\n")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- ncol(eta) / M1 # extra$NOS
if (NOS != 1) stop("can only handle 1 response")
is.a.mixed <- if (tmp3.TF[ 2])
rowSums(extra$skip.mix.a) > 0 else rep(FALSE, n)
is.i.mixed <- if (tmp3.TF[ 4])
rowSums(extra$skip.mix.i) > 0 else rep(FALSE, n)
is.d.mixed <- if (tmp3.TF[ 6])
rowSums(extra$skip.mix.d) > 0 else rep(FALSE, n)
is.a.mlmed <- if (tmp3.TF[ 8])
rowSums(extra$skip.mlm.a) > 0 else rep(FALSE, n)
is.i.mlmed <- if (tmp3.TF[ 9])
rowSums(extra$skip.mlm.i) > 0 else rep(FALSE, n)
is.d.mlmed <- if (tmp3.TF[10])
rowSums(extra$skip.mlm.d) > 0 else rep(FALSE, n)
is.ns <- !is.a.mlmed & !is.i.mlmed & !is.d.mlmed &
!is.a.mixed & !is.i.mixed & !is.d.mixed # & !is.truncd
prob.mlm.a <- if (la.mlm) rowSums(pobs.mlm) else 0 # scalar okay
prob.mlm.i <- if (li.mlm) rowSums(pstr.mlm) else 0 # scalar okay
prob.mlm.d <- if (ld.mlm) rowSums(pdip.mlm) else 0 # scalar okay
pmf.deriv1 <- function(y, shape) {
-dzeta(y, shape) * (log(y) +
zeta(shape + 1, deriv = 1) / zeta(shape + 1))
}
pmf.deriv2 <- function(y, shape) {
tmp2 <- zeta(shape + 1, deriv = 1) / zeta(shape + 1)
dzeta(y, shape) * ((log(y) + tmp2)^2 -
zeta(shape + 1, deriv = 2) / zeta(shape + 1) + tmp2^2)
}
sumD.mix.1a.p <- sumD.mix.2a.p <- matrix(0, n, NOS)
if (la.mix > 0) { # \calA_p
DA.mix.0mat.a <- # Matches naming convention further below
DA.mix.1mat.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
sumD.mix.1a.p <- sumD.mix.1a.p + pmf.deriv1(aval, shape.p)
sumD.mix.2a.p <- sumD.mix.2a.p + pmf.deriv2(aval, shape.p)
pmf.a <- dzeta(aval, shape.a)
DA.mix.0mat.a[, jay] <- pmf.a
DA.mix.1mat.a[, jay] <- pmf.deriv1(aval, shape.a)
}
Denom1.a <- rowSums(DA.mix.1mat.a) # aka sumD.mix.1a.a
} # la.mix > 0
if (li.mix) {
DI.mix.0mat.i <- # wrt inflated distribution
DI.mix.1mat.i <- DI.mix.2mat.i <- matrix(0, n, li.mix)
DP.mix.0mat.i <- # wrt parent distribution
DP.mix.1mat.i <- DP.mix.2mat.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.i <- dzeta(ival, shape.i)
DI.mix.0mat.i[, jay] <- pmf.i
DI.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.i)
DI.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.i)
pmf.p <- dzeta(ival, shape.p)
DP.mix.0mat.i[, jay] <- pmf.p
DP.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.p)
DP.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
Denom1.i <- rowSums(DI.mix.1mat.i)
Denom2.i <- rowSums(DI.mix.2mat.i)
} # li.mix
if (ld.mix) {
DD.mix.0mat.d <- # wrt deflated distribution
DD.mix.1mat.d <- DD.mix.2mat.d <- matrix(0, n, ld.mix)
DP.mix.0mat.d <- # wrt parent distribution
DP.mix.1mat.d <- DP.mix.2mat.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.d <- dzeta(dval, shape.d)
DD.mix.0mat.d[, jay] <- pmf.d
DD.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.d)
DD.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.d)
pmf.p <- dzeta(dval, shape.p)
DP.mix.0mat.d[, jay] <- pmf.p
DP.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.p)
DP.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
Denom1.d <- rowSums(DD.mix.1mat.d)
Denom2.d <- rowSums(DD.mix.2mat.d)
} # ld.mix
Bits <- moments.gaitdcombo.zeta(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
sumD.mlm.1a.p <- sumD.mlm.2a.p <- matrix(0, n, NOS)
if (la.mlm)
for (aval in a.mlm) {
sumD.mlm.1a.p <- sumD.mlm.1a.p + pmf.deriv1(aval, shape.p)
sumD.mlm.2a.p <- sumD.mlm.2a.p + pmf.deriv2(aval, shape.p)
}
Denom0.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
Numer <- 1 - pobs.mix - pstr.mix - prob.mlm.a - prob.mlm.i +
pdip.mix + prob.mlm.d
Denom0.a <- c(Bits[["SumA0.mix.a"]]) # Not .p
Denom0.i <- c(Bits[["SumI0.mix.i"]])
Denom0.d <- c(Bits[["SumD0.mix.d"]])
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, NOS)
if (li.mlm > 0) {
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
pmf.p <- dzeta(ival, shape.p)
Dp.mlm.0Mat.i[, jay] <- pmf.p
Dp.mlm.1Mat.i[, jay] <- pmf.deriv1(ival, shape.p)
Dp.mlm.2Mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
} # li.mlm
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, NOS)
if (ld.mlm > 0) {
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
pmf.p <- dzeta(dval, shape.p)
Dp.mlm.0Mat.d[, jay] <- pmf.p
Dp.mlm.1Mat.d[, jay] <- pmf.deriv1(dval, shape.p)
Dp.mlm.2Mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
} # ld.mlm
sumD.1t.p <- sumD.2t.p <-
sumD.1t.a <- sumD.2t.a <-
sumD.1t.i <- sumD.2t.i <-
sumD.1t.d <- sumD.2t.d <- matrix(0, n, NOS)
if (ltruncat)
for (tval in truncate) {
sumD.1t.p <- sumD.1t.p + pmf.deriv1(tval, shape.p)
sumD.2t.p <- sumD.2t.p + pmf.deriv2(tval, shape.p)
sumD.1t.a <- sumD.1t.a + pmf.deriv1(tval, shape.a)
sumD.2t.a <- sumD.2t.a + pmf.deriv2(tval, shape.a)
sumD.1t.i <- sumD.1t.i + pmf.deriv1(tval, shape.i)
sumD.2t.i <- sumD.2t.i + pmf.deriv2(tval, shape.i)
sumD.1t.d <- sumD.1t.d + pmf.deriv1(tval, shape.d)
sumD.2t.d <- sumD.2t.d + pmf.deriv2(tval, shape.d)
}
if (is.finite(max.support)) {
stop("argument 'max.support' must be 'Inf'.")
} # is.finite(max.support)
Denom1.p <- c(-sumD.1t.p - sumD.mlm.1a.p - sumD.mix.1a.p)
Denom2.p <- c(-sumD.2t.p - sumD.mlm.2a.p - sumD.mix.2a.p)
d0B.PI.mlm <- Dp.mlm.0Mat.i / Denom0.p
d1B.PI.mlm <- Dp.mlm.1Mat.i / Denom0.p - # This is most general
Dp.mlm.0Mat.i * Denom1.p / Denom0.p^2
d2B.PI.mlm <- Dp.mlm.2Mat.i / Denom0.p -
2 * Dp.mlm.1Mat.i * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.i * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.i * (Denom1.p^2) / Denom0.p^3
d0B.PD.mlm <- Dp.mlm.0Mat.d / Denom0.p
d1B.PD.mlm <- Dp.mlm.1Mat.d / Denom0.p - # This is most general
Dp.mlm.0Mat.d * Denom1.p / Denom0.p^2
d2B.PD.mlm <- Dp.mlm.2Mat.d / Denom0.p -
2 * Dp.mlm.1Mat.d * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.d * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.d * (Denom1.p^2) / Denom0.p^3
DELTA.i.mlm <- if (li.mlm > 0) {
Numer * d0B.PI.mlm + pstr.mlm # n x li.mlm.
} else {
matrix(0, n, 1) # If li.mlm == 0, for rowSums().
}
DELTA.d.mlm <- if (ld.mlm > 0) {
Numer * d0B.PD.mlm - pdip.mlm # n x ld.mlm.
} else {
matrix(0, n, 1) # If ld.mlm == 0, for rowSums().
}
if (li.mix > 0) {
d0A.i <- DI.mix.0mat.i / Denom0.i
d0B.PI.mix <- DP.mix.0mat.i / Denom0.p
DELTA.i.mix <- Numer * d0B.PI.mix + pstr.mix * d0A.i
d1A.i <- (DI.mix.1mat.i - DI.mix.0mat.i *
Denom1.i / Denom0.i) / Denom0.i
d2A.i <- (DI.mix.2mat.i - (2 * DI.mix.1mat.i * Denom1.i +
DI.mix.0mat.i * Denom2.i) / Denom0.i +
2 * DI.mix.0mat.i * (Denom1.i / Denom0.i)^2) / Denom0.i
d1B.PI.mix <- DP.mix.1mat.i / Denom0.p -
DP.mix.0mat.i * Denom1.p / Denom0.p^2
d2B.PI.mix <- DP.mix.2mat.i / Denom0.p -
2 * DP.mix.1mat.i * Denom1.p / Denom0.p^2 -
DP.mix.0mat.i * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.i * (Denom1.p^2) / Denom0.p^3
} # li.mix > 0
if (ld.mix > 0) {
d0A.d <- DD.mix.0mat.d / Denom0.d
d0B.PD.mix <- DP.mix.0mat.d / Denom0.p
DELTA.d.mix <- Numer * d0B.PD.mix - pdip.mix * d0A.d
d1A.d <- (DD.mix.1mat.d - DD.mix.0mat.d *
Denom1.d / Denom0.d) / Denom0.d
d2A.d <- (DD.mix.2mat.d - (2 * DD.mix.1mat.d * Denom1.d +
DD.mix.0mat.d * Denom2.d) / Denom0.d +
2 * DD.mix.0mat.d * (Denom1.d / Denom0.d)^2) / Denom0.d
d1B.PD.mix <- DP.mix.1mat.d / Denom0.p -
DP.mix.0mat.d * Denom1.p / Denom0.p^2
d2B.PD.mix <- DP.mix.2mat.d / Denom0.p -
2 * DP.mix.1mat.d * Denom1.p / Denom0.p^2 -
DP.mix.0mat.d * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.d * (Denom1.p^2) / Denom0.p^3
} # ld.mix > 0
if (la.mix) {
d0A.a <- DA.mix.0mat.a / Denom0.a
d1A.a <- DA.mix.1mat.a / Denom0.a -
DA.mix.0mat.a * Denom1.a / Denom0.a^2
} # la.mix
fred0.p <- zeta(shape.p + 1)
fred1.p <- zeta(shape.p + 1, deriv = 1)
dl.dshape.p <- -log(y) - fred1.p / fred0.p # Usual formula
dl.dshape.p[!is.ns] <- 0 # For is.a.mixed & is.a.mlmed
dl.dshape.a <- dl.dshape.i <- dl.dshape.d <- numeric(n)
dl.dpstr.mix <- (-1) / Numer # \notin A, I, T, D
dl.dpstr.mix[is.a.mixed] <- 0
dl.dpstr.mix[is.a.mlmed] <- 0
dl.dpdip.mix <- (+1) / Numer # \notin A, I, T, D
dl.dpdip.mix[is.a.mixed] <- 0
dl.dpdip.mix[is.a.mlmed] <- 0
dl.dpobs.mix <- numeric(n) # 0 for \calA_{np}
dl.dpobs.mix[is.ns] <- (-1) / Numer[is.ns]
dl.dpobs.mlm <-
dl.dpstr.mlm <- matrix(0, n, 1) # May be unneeded
dl.dpdip.mlm <- matrix(0, n, max(1, ld.mlm)) # Initzed if used.
dl.dpdip.mlm[is.ns, ] <- 1 / Numer[is.ns]
if (tmp3.TF[ 8] && la.mlm) { # aka \calA_{np}
dl.dpobs.mlm <- matrix(-1 / Numer, n, la.mlm) # \notin calS
dl.dpobs.mlm[!is.ns, ] <- 0 # For a.mix only really
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
is.alt.j.mlm <- extra$skip.mlm.a[, jay] # Logical vector
tmp7a <- 1 / pobs.mlm[is.alt.j.mlm, jay]
dl.dpobs.mlm[is.alt.j.mlm, jay] <- tmp7a
} # jay
} # la.mlm
dl.dshape.p[is.ns] <- dl.dshape.p[is.ns] -
(Denom1.p / Denom0.p)[is.ns]
if (tmp3.TF[ 9] && li.mlm > 0) { # aka \calI_{np}
dl.dpstr.mlm <- matrix(-1 / Numer, n, li.mlm)
dl.dpstr.mlm[!is.ns, ] <- 0 # For a.mlm and a.mix
for (jay in seq(li.mlm)) {
is.inf.j.mlm <- extra$skip.mlm.i[, jay] # Logical vector
tmp7i <- Numer * d1B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
dl.dshape.p[is.inf.j.mlm] <- tmp7i[is.inf.j.mlm]
tmp9i <- d0B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
n.tmp <- -tmp9i[is.inf.j.mlm]
p.tmp <- +tmp9i[is.inf.j.mlm]
if (tmp3.TF[ 8] && la.mlm)
dl.dpobs.mlm[is.inf.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix)
dl.dpobs.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[ 4] && li.mix)
dl.dpstr.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[10] && ld.mlm)
dl.dpdip.mlm[is.inf.j.mlm, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix)
dl.dpdip.mix[is.inf.j.mlm ] <- p.tmp
tmp8 <- (1 - d0B.PI.mlm[, jay]) / DELTA.i.mlm[, jay]
dl.dpstr.mlm[is.inf.j.mlm, ] <- n.tmp # tmp9[is.inf.j.mlm]
dl.dpstr.mlm[is.inf.j.mlm, jay] <- tmp8[is.inf.j.mlm]
} # jay
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # aka \calD_{np}
for (jay in seq(ld.mlm)) {
is.def.j.mlm <- extra$skip.mlm.d[, jay] # Logical vector
tmp7d <- Numer * d1B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
dl.dshape.p[is.def.j.mlm] <- tmp7d[is.def.j.mlm] # 20211020
tmp9d <- d0B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
p.tmp <- +tmp9d[is.def.j.mlm]
n.tmp <- -tmp9d[is.def.j.mlm]
if (tmp3.TF[ 9] && li.mlm)
dl.dpstr.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix)
dl.dpstr.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm)
dl.dpobs.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix)
dl.dpobs.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 6] && ld.mix)
dl.dpdip.mix[is.def.j.mlm ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, jay] <-
dl.dpdip.mlm[is.def.j.mlm, jay] -
1 / DELTA.d.mlm[is.def.j.mlm, jay]
} # jay
} # ld.mlm > 0
if (tmp3.TF[ 2] && la.mix) { # aka \calA_{p}
dl.dpobs.mix[is.a.mixed] <- 1 / pobs.mix[is.a.mixed]
if (tmp3.TF[ 3] && la.mix > 1)
for (jay in seq(la.mix)) {
is.alt.j.mix <- extra$skip.mix.a[, jay] # Logical vector
tmp2 <- d1A.a[, jay] / d0A.a[, jay]
dl.dshape.a[is.alt.j.mix] <- tmp2[is.alt.j.mix] # ccc.
} # jay
} # la.mix
if (tmp3.TF[ 4] && li.mix > 0) { # aka \calI_{p}
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
is.inf.j.mix <- extra$skip.mix.i[, jay] # Logical vector
tmp7b <- Numer * d1B.PI.mix[, jay] / DELTA.i.mix[, jay]
dl.dshape.p[is.inf.j.mix] <- tmp7b[is.inf.j.mix]
tmp8 <- (d0A.i[, jay] - d0B.PI.mix[, jay]) / DELTA.i.mix[, jay]
dl.dpstr.mix[is.inf.j.mix] <- tmp8[is.inf.j.mix]
if (li.mix > 1) {
tmp2 <- pstr.mix * d1A.i[, jay] / DELTA.i.mix[, jay]
dl.dshape.i[is.inf.j.mix] <- tmp2[is.inf.j.mix]
}
tmp9i <- d0B.PI.mix[, jay] / DELTA.i.mix[, jay]
n.tmp <- -tmp9i[is.inf.j.mix]
p.tmp <- +tmp9i[is.inf.j.mix]
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mix, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mix ] <- p.tmp
} # jay
} # li.mix > 0
if (tmp3.TF[ 6] && ld.mix > 0) { # aka \calD_{p}
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
is.def.j.mix <- extra$skip.mix.d[, jay] # Logical vector
tmp7b <- Numer * d1B.PD.mix[, jay] / DELTA.d.mix[, jay]
dl.dshape.p[is.def.j.mix] <- tmp7b[is.def.j.mix]
tmp8 <- (d0B.PD.mix[, jay] - d0A.d[, jay]) / DELTA.d.mix[, jay]
dl.dpdip.mix[is.def.j.mix] <- tmp8[is.def.j.mix]
if (ld.mix > 1) {
tmp2 <- (-pdip.mix) * d1A.d[, jay] / DELTA.d.mix[, jay]
dl.dshape.d[is.def.j.mix] <- tmp2[is.def.j.mix]
}
tmp9d <- d0B.PD.mix[, jay] / DELTA.d.mix[, jay]
n.tmp <- -tmp9d[is.def.j.mix]
p.tmp <- +tmp9d[is.def.j.mix]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.def.j.mix, ] <- p.tmp
} # jay
} # ld.mix > 0
new.ansd <- matrix(0, n, M) # Same dimension as eta
tmp3.TF <- !is.na(rowSums(extra$indeta))
if (lall.len) { # An MLM fitted
all6.dldp <- cbind(if (tmp3.TF[ 2]) dl.dpobs.mix else NULL,
if (tmp3.TF[ 4]) dl.dpstr.mix else NULL,
if (tmp3.TF[ 6]) dl.dpdip.mix else NULL,
if (tmp3.TF[ 8]) dl.dpobs.mlm else NULL,
if (tmp3.TF[ 9]) dl.dpstr.mlm else NULL,
if (tmp3.TF[10]) dl.dpdip.mlm else NULL)
rSs.tmp <- rowSums(allprobs[, -ncol(allprobs), drop = FALSE] *
all6.dldp)
new.ansd[, -ind.shape.z] <- allprobs[, -ncol(allprobs)] *
(all6.dldp - rSs.tmp)
} # lall.len
dshape.p.deta <- dtheta.deta(shape.p, .lshape.p , .eshape.p )
if (tmp3.TF[ 3])
dshape.a.deta <- dtheta.deta(shape.a, .lshape.a , .eshape.a )
if (tmp3.TF[ 5])
dshape.i.deta <- dtheta.deta(shape.i, .lshape.i , .eshape.i )
if (tmp3.TF[ 7])
dshape.d.deta <- dtheta.deta(shape.d, .lshape.d , .eshape.d )
new.ansd[, 1] <- dl.dshape.p * dshape.p.deta
if (tmp3.TF[ 3])
new.ansd[, extra$indeta[3, 1]] <- dl.dshape.a * dshape.a.deta
if (tmp3.TF[ 5])
new.ansd[, extra$indeta[5, 1]] <- dl.dshape.i * dshape.i.deta
if (tmp3.TF[ 7])
new.ansd[, extra$indeta[7, 1]] <- dl.dshape.d * dshape.d.deta
onecoln.indeta <- extra$indeta[1:7, ] # One coln params only
onecoln.indeta <- na.omit(onecoln.indeta) # Only those present
allcnames <- c(rownames(onecoln.indeta),
as.character(c(a.mlm, i.mlm, d.mlm)))
colnames(new.ansd) <- allcnames
c(w) * new.ansd
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.truncate = truncate, .max.support = max.support ))),
weight = eval(substitute(expression({ # gaitdzeta
wz <- matrix(0, n, M * (M + 1) / 2) # The complete size
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom0.p)
if (min(probns) < 0 || 1 < max(probns))
stop("variable 'probns' for P(nonspecial) is out of range")
zero0n <- numeric(n)
ned2l.dpobs.mix.shape.p <- zero0n # mB overwritten below [4279]
ned2l.dpobs.mix.shape.a <- zero0n # Fini; (2, 3) element
ned2l.dpobs.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpobs.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.p <- zero0n # Optional (1, 4) element
ned2l.dpstr.mix.shape.a <- zero0n # Final; nothing to do
ned2l.dpstr.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpdip.mix.shape.p <- zero0n # Optional (1, 6) element
posn.pobs.mix <- as.vector(extra$indeta[ 2, 'launch'])
posn.shape.a <- as.vector(extra$indeta[ 3, 'launch'])
posn.pstr.mix <- as.vector(extra$indeta[ 4, 'launch'])
posn.shape.i <- as.vector(extra$indeta[ 5, 'launch'])
posn.pdip.mix <- as.vector(extra$indeta[ 6, 'launch'])
posn.shape.d <- as.vector(extra$indeta[ 7, 'launch'])
posn.pobs.mlm <- as.vector(extra$indeta[ 8, 'launch'])
posn.pstr.mlm <- as.vector(extra$indeta[ 9, 'launch'])
posn.pdip.mlm <- as.vector(extra$indeta[10, 'launch'])
ned2l.dpdip.mix2 <- # Elt (6, 6)
ned2l.dpstr.mix2 <- # Elt (4, 4). Unchanged by deflation.
ned2l.dpobs.mlm.pstr.mix <- # Elts (4, >=8). (((09)))
ned2l.dpobs.mix.pstr.mix <- +probns / Numer^2 # ccc Elt (2, 4)
if (all(c(la.mix, li.mlm) > 0)) # (((08)))
ned2l.dpobs.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(li.mix, li.mlm) > 0)) # (((10)))
ned2l.dpstr.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(ld.mix, ld.mlm) > 0)) # (((21)))
ned2l.dpdip.mix.pdip.mlm <- matrix( probns / Numer^2, n, ld.mlm)
ned2l.dpobs.mlm.pdip.mix <- # Elts (6, >=8). (((19)))
ned2l.dpstr.mix.pdip.mix <- # Elt (4, 6)
ned2l.dpobs.mix.pdip.mix <- -probns / Numer^2 # ccc Elt (2, 6)
if (all(c(la.mix, ld.mlm) > 0)) # (((17)))
ned2l.dpobs.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(li.mix, ld.mlm) > 0)) # (((18)))
ned2l.dpstr.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(ld.mix, li.mlm) > 0)) # (((20)))
ned2l.dpdip.mix.pstr.mlm <- matrix(-probns / Numer^2, n, li.mlm)
ned2l.dshape.p2 <- probns * (
zeta(shape.p + 1, deriv = 2) / fred0.p -
(fred1.p / fred0.p)^2 + # ccc
Denom2.p / Denom0.p - (Denom1.p / Denom0.p)^2) +
(if (tmp3.TF[ 4] && li.mix) Numer *
rowSums(Numer * (d1B.PI.mix^2) / DELTA.i.mix - d2B.PI.mix) else 0) +
(if (tmp3.TF[ 9] && li.mlm) Numer *
rowSums(Numer * (d1B.PI.mlm^2) / DELTA.i.mlm - d2B.PI.mlm) else 0) +
(if (tmp3.TF[ 6] && ld.mix) Numer *
rowSums(Numer * (d1B.PD.mix^2) / DELTA.d.mix - d2B.PD.mix) else 0) +
(if (tmp3.TF[10] && ld.mlm) Numer * # nnn.
rowSums(Numer * (d1B.PD.mlm^2) / DELTA.d.mlm - d2B.PD.mlm) else 0)
wz[, iam(1, 1, M)] <- ned2l.dshape.p2 * dshape.p.deta^2
ned2l.dpobs.mix2 <- 1 / pobs.mix + probns / Numer^2
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpobs.mix2 <- # More just below, ccc
ned2l.dpobs.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpobs.mix2 <- # ccc.
ned2l.dpobs.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (tmp3.TF[ 6] && ld.mix > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (tmp3.TF[ 2] && la.mix > 0)
wz[, iam(2, 2, M)] <- ned2l.dpobs.mix2 # Link done later
if (tmp3.TF[ 3] && la.mix > 1) {
ned2l.dshape.a2 <- pobs.mix * (
rowSums((DA.mix.1mat.a^2) / DA.mix.0mat.a) / Denom0.a -
(Denom1.a / Denom0.a)^2) # ccc.
wz[, iam(3, 3, M)] <- ned2l.dshape.a2 * dshape.a.deta^2
}
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 +
rowSums((d0A.i - d0B.PI.mix)^2 / DELTA.i.mix)
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mix * (1 + Numer * (d0A.i - d0B.PI.mix) / DELTA.i.mix))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (all(tmp3.TF[c(2, 4)]))
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix +
rowSums(-d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
} # (tmp3.TF[ 4] && li.mix > 0)
if (all(tmp3.TF[c(2, 4, 9)])) { # was la.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 4, 6)])) { # == ld.mix > 0 & DELTA.d.mix
ned2l.dpobs.mix.pstr.mix <- # nnn
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(2, 4, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pstr.mix <- # nnn.
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pstr.mix))
wz[, iam(posn.pobs.mix, posn.pstr.mix, M)] <-
ned2l.dpobs.mix.pstr.mix # Link done later
if (all(tmp3.TF[c(2, 6)]))
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix +
rowSums( d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
if (all(tmp3.TF[c(2, 6, 9)])) { # == li.mlm > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 6, 4)])) { # == li.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (all(tmp3.TF[c(2, 6, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pdip.mix <- # nnn.
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pdip.mix))
wz[, iam(posn.pobs.mix, posn.pdip.mix, M)] <-
ned2l.dpobs.mix.pdip.mix # Link done later
if (tmp3.TF[ 5] && li.mix > 1) { # \calI_{p}, includes \theta_i.
ned2l.dshape.p.shape.i <- pstr.mix * Numer *
rowSums(d1A.i * d1B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(1, posn.shape.i, M)] <- ned2l.dshape.p.shape.i *
dshape.p.deta * dshape.i.deta # All links done here
ned2l.dshape.i2 <- pstr.mix *
rowSums(pstr.mix * (d1A.i^2) / DELTA.i.mix - d2A.i) # ccc.
wz[, iam(posn.shape.i, posn.shape.i, M)] <-
ned2l.dshape.i2 * dshape.i.deta^2
if (tmp3.TF[ 2]) { # tmp3.TF[ 4] is TRUE, given tmp3.TF[ 5]
ned2l.dpobs.mix.shape.i <-
rowSums(-pstr.mix * d1A.i * d0B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(posn.pobs.mix, posn.shape.i, M)] <-
ned2l.dpobs.mix.shape.i # * dshape.i.deta done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.i <- rowSums( # ccc.
d1A.i * (pstr.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix - 1))
wz[, iam(posn.pstr.mix, posn.shape.i, M)] <-
ned2l.dpstr.mix.shape.i # * dshape.i.deta done later
}
if (all(tmp3.TF[c(5, 6)])) {
ned2l.dpdip.mix.shape.i <- rowSums(
(-pstr.mix) * d0B.PI.mix * d1A.i / DELTA.i.mix)
wz[, iam(posn.pdip.mix, posn.shape.i, M)] <-
ned2l.dpdip.mix.shape.i # link done later
}
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.i <- rowSums(
-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix) # ccc.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.i, M)] <-
ned2l.dpobs.mlm.shape.i # * dshape.i.deta done later
}
} # (tmp3.TF[ 5] && li.mix > 1)
if (tmp3.TF[ 6] && ld.mix > 0) { # \calD_{p}, maybe w. \theta_d
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mix * (1 + Numer * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 +
rowSums((d0A.d - d0B.PD.mix)^2 / DELTA.d.mix)
} # (tmp3.TF[ 6] && ld.mix > 0)
if (tmp3.TF[ 7] && ld.mix > 1) { # \calD_{p}, includes \theta_d
ned2l.dshape.p.shape.d <- (-pdip.mix) * Numer *
rowSums(d1A.d * d1B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(1, posn.shape.d, M)] <- ned2l.dshape.p.shape.d *
dshape.p.deta * dshape.d.deta # All links done here
if (tmp3.TF[ 2]) { # tmp3.TF[ 6] is TRUE, given tmp3.TF[ 7]
ned2l.dpobs.mix.shape.d <-
rowSums(pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(posn.pobs.mix, posn.shape.d, M)] <-
ned2l.dpobs.mix.shape.d # link done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.d <- rowSums(
pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix)
wz[, iam(posn.pstr.mix, posn.shape.d, M)] <-
ned2l.dpstr.mix.shape.d # * dshape.i.deta done later
}
ned2l.dpdip.mix.shape.d <- rowSums(
d1A.d * (1 + pdip.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
wz[, iam(posn.pdip.mix, posn.shape.d, M)] <-
ned2l.dpdip.mix.shape.d # * dshape.d.deta done later
ned2l.dshape.d2 <- pdip.mix *
rowSums(pdip.mix * (d1A.d^2) / DELTA.d.mix + d2A.d) # nnn.
wz[, iam(posn.shape.d, posn.shape.d, M)] <-
ned2l.dshape.d2 * dshape.d.deta^2
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.d <- rowSums(
pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix) # nnn.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.d, M)] <-
ned2l.dpobs.mlm.shape.d # * dshape.d.deta done later
}
} # (tmp3.TF[ 7] && ld.mix > 1)
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # ccc
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ned2l.dpstr.mix.shape.p <- # ccc.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
} # tmp3.TF[ 9] && li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # nnn.
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpstr.mix.shape.p <- # nnn.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
} # tmp3.TF[10] && ld.mlm > 0
if (!is.na(posn.pobs.mix)) # Optional (1, 2) element:
wz[, iam(1, posn.pobs.mix, M)] <-
ned2l.dpobs.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix)) # Optional (1, 4) element
wz[, iam(1, posn.pstr.mix, M)] <-
ned2l.dpstr.mix.shape.p # One link done later
if (!is.na(posn.pdip.mix)) # Optional (1, 6) element
wz[, iam(1, posn.pdip.mix, M)] <-
ned2l.dpdip.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix) &&
!is.na(posn.pdip.mix)) # Optional (4, 6) element
wz[, iam(posn.pstr.mix, posn.pdip.mix, M)] <-
ned2l.dpstr.mix.pdip.mix # Links done later zz1
if (!is.na(posn.pstr.mix)) # Optional (4, 4) element
wz[, iam(posn.pstr.mix, # Link done later
posn.pstr.mix, M)] <- ned2l.dpstr.mix2
if (!is.na(posn.pdip.mix)) # Optional (6, 6) element
wz[, iam(posn.pdip.mix, # Link done later
posn.pdip.mix, M)] <- ned2l.dpdip.mix2
if (tmp3.TF[ 8] && la.mlm) { # \calA_{np}, includes \omega_s
ofset <- posn.pobs.mlm - 1 # 7 for GAITD combo
for (uuu in seq(la.mlm)) { # Diagonal elts only
wz[, iam(ofset + uuu,
ofset + uuu, M)] <- 1 / pobs.mlm[, uuu]
} # uuu
tmp8a <- probns / Numer^2
if (tmp3.TF[ 4] && li.mix)
tmp8a <- tmp8a + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
tmp8a <- tmp8a + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
tmp8a <- tmp8a + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
tmp8a <- tmp8a + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # All elts
for (vvv in uuu:la.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
wz[, iam(ofset + uuu, ofset + vvv, M)] + tmp8a # All elts
} # la.mlm
if (tmp3.TF[ 8] && la.mlm) {
init0.i.val <- init0.d.val <- 0
if (tmp3.TF[ 9] && li.mlm) init0.i.val <-
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[10] && ld.mlm) init0.d.val <-
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpobs.mlm.shape.p <- init0.i.val + init0.d.val # Vector
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums( # nnn
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ofset <- posn.pobs.mlm - 1 # 5 for combo
for (vvv in seq(la.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpobs.mlm.shape.p
} # la.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s
init0.val <- probns / Numer^2
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (ld.mix) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (ld.mlm) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
ned2l.dpstr.mlm2 <-
matrix(init0.val, n, li.mlm * (li.mlm + 1) / 2)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss])^2 / DELTA.i.mlm[, sss]
if (li.mlm > 1) {
for (uuu in seq(li.mlm - 1))
for (vvv in (uuu + 1):li.mlm)
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
} # if (li.mlm > 1)
ofset <- posn.pstr.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in uuu:li.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s
init0.val <- probns / Numer^2
if (ld.mix)
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (li.mlm)
init0.val <- init0.val + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
ned2l.dpdip.mlm2 <-
matrix(init0.val, n, ld.mlm * (ld.mlm + 1) / 2)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu))^2 / DELTA.d.mlm[, sss]
if (ld.mlm > 1) {
for (uuu in seq(ld.mlm - 1))
for (vvv in (uuu + 1):ld.mlm)
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu)) *
(d0B.PD.mlm[, sss] - (sss == vvv)) / DELTA.d.mlm[, sss]
} # if (ld.mlm > 1)
ofset <- posn.pdip.mlm - 1
for (uuu in seq(ld.mlm))
for (vvv in uuu:ld.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)]
} # ld.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpstr.mlm.theta.p <- matrix(0, n, li.mlm)
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.theta.p[, vvv] <-
ned2l.dpstr.mlm.theta.p[, vvv] +
d1B.PI.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PI.mlm[, sss]) / (
DELTA.i.mlm[, sss]))
if (li.mix && tmp3.TF[ 4])
ned2l.dpstr.mlm.theta.p <-
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpstr.mlm.theta.p <- # nnn
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (ld.mlm && tmp3.TF[10])
ned2l.dpstr.mlm.theta.p <- # nnn.
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ofset <- posn.pstr.mlm - 1
for (vvv in seq(li.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpstr.mlm.theta.p[, vvv]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpdip.mlm.theta.p <- matrix(0, n, ld.mlm)
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm.theta.p[, vvv] <-
ned2l.dpdip.mlm.theta.p[, vvv] - # Minus
d1B.PD.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PD.mlm[, sss]) / (
DELTA.d.mlm[, sss]))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (li.mix && tmp3.TF[ 4])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (li.mlm && tmp3.TF[ 9])
ned2l.dpdip.mlm.theta.p <- # nnn.
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ofset <- posn.pdip.mlm - 1
for (vvv in seq(ld.mlm)) # nnn.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpdip.mlm.theta.p[, vvv]
} # ld.mlm > 0
if (li.mlm && li.mix > 1) {
ned2l.dpstr.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.i, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.i # ccc.
} # li.mlm && li.mix > 1
if (ld.mlm && ld.mix > 1) {
ned2l.dpdip.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.d, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.d # nnn.
} # ld.mlm && ld.mix > 1
if (ld.mlm && li.mix > 1) {
ned2l.dpdip.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.i, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.i # nnn.
} # ld.mlm && li.mix > 1
if (li.mlm && ld.mix > 1) {
ned2l.dpstr.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.d, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.d # nnn.
} # li.mlm && ld.mix > 1
if (all(c(la.mlm, li.mlm) > 0)) {
ned2l.dpobs.mlm.pstr.mlm <-
array(probns / Numer^2, c(n, la.mlm, li.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] - d0B.PI.mlm[, sss] *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pstr.mlm <-
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pstr <- posn.pstr.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pstr + vvv, M)] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(li.mlm, ld.mlm) > 0)) {
ned2l.dpstr.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, li.mlm, ld.mlm))
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PI.mlm[, sss] *
((sss == uuu) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpstr.mlm.pdip.mlm <-
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpstr.mlm.pdip.mlm <- # nnn.
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pstr <- posn.pstr.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pstr + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv]
} # all(c(li.mlm, ld.mlm) > 0)
if (all(c(la.mlm, ld.mlm) > 0)) {
ned2l.dpobs.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, la.mlm, ld.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(la.mix, la.mlm) > 0)) {
ned2l.dpobs.mix.pobs.mlm <- probns / Numer^2 # Initialize
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 7]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pobs.mix, posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pobs.mlm # Link done later
}
if (all(c(la.mix, li.mlm) > 0)) { # all(tmp3.TF[c(2, 9)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pstr.mlm <-
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pstr.mlm <- # nnn
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mix.pstr.mlm <- # nnn; + is correct, not -
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mix.pstr.mlm[, uuu] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm)) # ccc.
wz[, iam(posn.pobs.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] # Link done later
} # all(c(la.mix, li.mlm) > 0)
if (all(c(la.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(2, 10)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mix.pdip.mlm[, uuu] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
for (uuu in seq(ld.mlm)) # nnn.
wz[, iam(posn.pobs.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] # Link done later
} # all(c(la.mix, ld.mlm) > 0)
if (all(c(li.mix, la.mlm) > 0)) { # all(tmp3.TF[c(4, 8)])
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pstr.mix <-
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ned2l.dpobs.mlm.pstr.mix <- # tmp3.TF[ 4] && li.mix
ned2l.dpobs.mlm.pstr.mix -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pstr.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pstr.mix # Link done later
} # all(c(li.mix, la.mlm) > 0
if (all(c(ld.mix, la.mlm) > 0)) { # all(tmp3.TF[c(6, 8)])
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
ned2l.dpobs.mlm.pdip.mix <- # all(tmp3.TF[c(6, 8)])
ned2l.dpobs.mlm.pdip.mix +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
for (uuu in seq(la.mlm)) # nnn.
wz[, iam(posn.pdip.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pdip.mix # Link done later
} # all(c(ld.mix, la.mlm) > 0
if (all(c(li.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpstr.mix.pstr.mlm[, uuu] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
ned2l.dpstr.mix.pstr.mlm <-
ned2l.dpstr.mix.pstr.mlm -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] # Link done later
} # all(c(li.mix, li.mlm) > 0
if (all(c(ld.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(6, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[ 9]
for (sss in seq(ld.mlm))
ned2l.dpdip.mix.pdip.mlm[, uuu] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] -
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm -
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] # Link done later
} # all(c(ld.mix, ld.mlm) > 0
if (all(c(ld.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpdip.mix.pstr.mlm[, uuu] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] # Link done later
} # all(c(ld.mix, li.mlm) > 0
if (all(c(li.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(4, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[10]
for (sss in seq(ld.mlm))
ned2l.dpstr.mix.pdip.mlm[, uuu] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (li.mix) # tmp3.TF[ 4]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm +
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpstr.mix.pdip.mlm <- # nnn.
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] # Link done later
} # all(c(li.mix, ld.mlm) > 0)
if (lall.len) {
wz.6 <- matrix(0, n, M * (M + 1) / 2) # Or == 0 * wz
ind.rc <- setdiff(1:M, ind.shape.z) # Contiguous rows and
lind.rc <- length(ind.rc) # cols of the DAMLM
# Copy in the thetas values: the looping is overkill.
for (uuu in ind.shape.z)
for (sss in seq(M))
wz.6[, iam(uuu, sss, M)] <- wz[, iam(uuu, sss, M)]
speed.up <- intercept.only && (
length(offset) == 1 || all(offset[1] == offset))
IND.mlm <- iam(NA, NA, lind.rc, both = TRUE, diag = TRUE)
n.use <- if (speed.up) 2 else n # For sandwich.mlm
if (!length(extra$ind.wz.match)) {
Imat <- matrix(NA, lind.rc, lind.rc)
for (jay in seq(lind.rc)) {
iptr <- jay
for (kay in (ind.rc[jay]):M) {
if (!any(kay %in% ind.shape.z)) {
Imat[jay, iptr] <-
which(extra$index.M$row == ind.rc[jay] &
extra$index.M$col == kay)
iptr <- iptr + 1
} # if
} # kay
} # jay
ind.wz.match <- Imat[cbind(IND.mlm$row.ind,
IND.mlm$col.ind)]
extra$ind.wz.match <- ind.wz.match # Assign it once
} # !length(extra$ind.wz.match)
filling <- if (speed.up)
wz[1:n.use, extra$ind.wz.match, drop = FALSE] else
wz[, extra$ind.wz.match, drop = FALSE]
M.mlm <- lind.rc
if (is.null(extra$iamlist)) {
extra$iamlist <- iamlist <-
iam(NA, NA, M = M.mlm, both = TRUE)
if (M.mlm > 1) { # Offdiagonal elts
extra$iamlist.nod <- iamlist.nod <-
iam(NA, NA, M.mlm, both = TRUE, diag = FALSE)
}
} # is.null(extra$iamlist)
iamlist <- extra$iamlist
iamlist.nod <- extra$iamlist.nod
MM12.mlm <- M.mlm * (M.mlm + 1) / 2
Qf3 <- rowSums(filling[, 1:M.mlm, drop = FALSE] * # Diag elts
(allprobs[1:n.use, 1:M.mlm, drop = FALSE])^2)
if (M.mlm > 1) # Offdiagonal elts
Qf3 <- Qf3 + 2 * rowSums(allprobs[1:n.use, iamlist.nod$row] *
filling[, -(1:M.mlm), drop = FALSE] * # n-vector
allprobs[1:n.use, iamlist.nod$col])
Qf3 <- matrix(Qf3, n.use, MM12.mlm)
Qf2rowsums <- matrix(0, n.use, M.mlm) # rowsums stored colwise
for (want in seq(M.mlm)) { # Want the \equiv of rowSums(Qf2a)
iamvec <- iam(want, 1:M.mlm, M = M.mlm) # Diagonals included
Qf2rowsums[, want] <-
rowSums(filling[, iamvec, drop = FALSE] *
allprobs[1:n.use, 1:M.mlm])
} # want
Qf2a <- Qf2rowsums[, iamlist$row]
Qf2b <- Qf2rowsums[, iamlist$col]
Qform <- filling - Qf2a - Qf2b + Qf3 # n x MM12.mlm
Qform <- Qform *
allprobs[1:n.use, iamlist$row, drop = FALSE] *
allprobs[1:n.use, iamlist$col, drop = FALSE]
wz.6[, extra$ind.wz.match] <- if (speed.up)
matrix(Qform[1, ], n, ncol(Qform), byrow = TRUE) else c(Qform)
dstar.deta <- cbind(dshape.p.deta,
if (tmp3.TF[ 3]) dshape.a.deta else NULL,
if (tmp3.TF[ 5]) dshape.i.deta else NULL,
if (tmp3.TF[ 7]) dshape.d.deta else NULL)
iptr <- 0
if (length(ind.shape.z))
for (uuu in ind.shape.z) { # Could delete 3 4 shape.a (orthog)
iptr <- iptr + 1
for (ttt in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <- 0 # Initialize
for (sss in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <-
wz.6[, iam(uuu, ind.rc[ttt], M)] +
allprobs[, sss] * (max(0, sss == ttt) -
allprobs[, ttt]) *
wz[, iam(uuu, ind.rc[sss], M)] * dstar.deta[, iptr]
} # sss
} # ttt
} # uuu
wz <- wz.6 # Completed
} # lall.len
if (lall.len) { # A MLM was fitted
mytiny <- (allprobs < sqrt(.Machine$double.eps)) |
(allprobs > 1.0 - sqrt(.Machine$double.eps))
atiny <- rowSums(mytiny) > 0
if (any(atiny)) {
ind.diags <- setdiff(1:M, ind.shape.z) # Exclude thetas
wz[atiny, ind.diags] <- .Machine$double.eps +
wz[atiny, ind.diags] * (1 + .Machine$double.eps^0.5)
}
} # lall.len
c(w) * wz
}), list( .truncate = truncate ))))
} # gaitdzeta
gaitdlog <-
function(a.mix = NULL, i.mix = NULL,
d.mix = NULL,
a.mlm = NULL, i.mlm = NULL, # Unstructured probs are
d.mlm = NULL, # contiguous
truncate = NULL, max.support = Inf,
zero = c("pobs", "pstr", "pdip"), # Pruned, handles all 6
eq.ap = TRUE, eq.ip = TRUE, eq.dp = TRUE,
parallel.a = FALSE, parallel.i = FALSE,
parallel.d = FALSE,
lshape.p = "logitlink",
lshape.a = lshape.p, # "logitlink", 20201117
lshape.i = lshape.p, # "logitlink", 20201117
lshape.d = lshape.p, # "logitlink", 20211011
type.fitted = c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"),
gshape.p = -expm1(-7 * ppoints(12)),
gpstr.mix = ppoints(7) / 3, # ppoints(9) / 2,
gpstr.mlm = ppoints(7) / (3 + length(i.mlm)),
imethod = 1,
mux.init = c(0.75, 0.5, 0.75), # Order is A, I, D.
ishape.p = NULL, ishape.a = ishape.p,
ishape.i = ishape.p, ishape.d = ishape.p,
ipobs.mix = NULL, ipstr.mix = NULL, # 0.25,
ipdip.mix = NULL, # 0.01, # Easy but inflexible 0.01
ipobs.mlm = NULL, ipstr.mlm = NULL, # 0.25,
ipdip.mlm = NULL, # 0.01, # NULL, Easy but inflexible
byrow.aid = FALSE,
ishrinkage = 0.95,
probs.y = 0.35) {
mux.init <- rep_len(mux.init, 3)
if (length(a.mix) == 0) a.mix <- NULL
if (length(i.mix) == 0) i.mix <- NULL
if (length(d.mix) == 0) d.mix <- NULL
if (length(a.mlm) == 0) a.mlm <- NULL
if (length(i.mlm) == 0) i.mlm <- NULL
if (length(d.mlm) == 0) d.mlm <- NULL
if (length(truncate) == 0) truncate <- NULL
lowsup <- 1
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support,
min.support = lowsup)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltruncat <- length(truncate <- sort(truncate))
ltrunc.use <- ltruncat > 0 || !is.infinite(max.support)
lshape.p <- as.list(substitute(lshape.p))
eshape.p <- link2list(lshape.p)
lshape.p <- attr(eshape.p, "function.name")
lshape.p.save <- lshape.p
lpobs.mix <- "multilogitlink" # \omega_p
epobs.mix <- list() # zz NULL for now 20200907 coz 'multilogitlink'
eshape.a <- link2list(lshape.a)
lshape.a <- attr(eshape.a, "function.name")
lpstr.mix <- "multilogitlink" # \phi_p
epstr.mix <- list() # zz NULL for now 20200907 coz 'multilogitlink'
lpdip.mix <- "multilogitlink" # zz unsure 20211002
epdip.mix <- list() # zz unsure 20211002
eshape.i <- link2list(lshape.i)
lshape.i <- attr(eshape.i, "function.name")
eshape.d <- link2list(lshape.d)
lshape.d <- attr(eshape.d, "function.name")
lshape.p.save <- lshape.p
gshape.p.save <- gshape.p
if (is.vector(zero) && is.character(zero) && length(zero) == 3) {
if (li.mix + li.mlm == 0)
zero <- setdiff(zero, "pstr")
if (la.mix + la.mlm == 0)
zero <- setdiff(zero, "pobs")
if (ld.mix + ld.mlm == 0)
zero <- setdiff(zero, "pdip")
if (length(zero) == 0)
zero <- NULL # Better than character(0)
}
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
if (lall.len + ltruncat == 0 && is.infinite(max.support))
return(eval(substitute(
logff(lshape = .lshape.p.save ,
gshape = .gshape.p.save ,
zero = NULL),
list( .lshape.p.save = lshape.p.save,
.gshape.p.save = gshape.p.save ))))
if (!is.logical(eq.ap) || length(eq.ap) != 1)
stop("argument 'eq.ap' must be a single logical")
if (!is.logical(eq.ip) || length(eq.ip) != 1)
stop("argument 'eq.ip' must be a single logical")
if (!is.logical(parallel.a) || length(parallel.a) != 1)
stop("argument 'parallel.a' must be a single logical")
if (!is.logical(parallel.i) || length(parallel.i) != 1)
stop("argument 'parallel.i' must be a single logical")
if (!is.logical(parallel.d) || length(parallel.d) != 1)
stop("argument 'parallel.d' must be a single logical")
if (FALSE) { # Comment this out to allow default eq.ap = TRUE, etc.
if (la.mix <= 1 && eq.ap)
stop("<= one unstructured altered value (no 'shape.a')",
", so setting 'eq.ap = TRUE' is meaningless")
if (li.mix <= 1 && eq.ip)
stop("<= one unstructured inflated value (no 'shape.i')",
", so setting 'eq.ip = TRUE' is meaningless")
if (ld.mix <= 1 && eq.dp)
stop("<= one unstructured deflated value (no 'shape.d')",
", so setting 'eq.dp = TRUE' is meaningless")
if (la.mlm <= 1 && parallel.a) # Only \omega_1
stop("<= one altered mixture probability, 'pobs", a.mlm,
"', so setting 'parallel.a = TRUE' is meaningless")
if (li.mlm <= 1 && parallel.i) # Only \phi_1
stop("<= one inflated mixture probability, 'pstr", i.mlm,
"', so setting 'parallel.i = TRUE' is meaningless")
if (ld.mlm <= 1 && parallel.d) # Only \psi_1
stop("<= one deflated mixture probability, 'pdip", d.mlm,
"', so setting 'parallel.d = TRUE' is meaningless")
} # FALSE
type.fitted.choices <-
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s")
type.fitted <- match.arg(type.fitted[1], type.fitted.choices)[1]
tmp7a <- if (la.mlm) paste0("pobs.mlm", a.mlm) else NULL
tmp7b <- if (li.mlm) paste0("pstr.mlm", i.mlm) else NULL
tmp7c <- if (ld.mlm) paste0("pdip.mlm", d.mlm) else NULL
tmp3 <- c(shape.p = lshape.p,
pobs.mix = if (la.mix) "multilogitlink" else NULL,
shape.a = if (la.mix > 1) lshape.a else NULL,
pstr.mix = if (li.mix) "multilogitlink" else NULL,
shape.i = if (li.mix > 1) lshape.i else NULL,
pdip.mix = if (ld.mix) "multilogitlink" else NULL,
shape.d = if (ld.mix > 1) lshape.d else NULL,
if (la.mlm) rep("multilogitlink", la.mlm) else NULL,
if (li.mlm) rep("multilogitlink", li.mlm) else NULL,
if (ld.mlm) rep("multilogitlink", ld.mlm) else NULL)
Ltmp3 <- length(tmp3)
if (la.mlm + li.mlm + ld.mlm)
names(tmp3)[(Ltmp3 - la.mlm - li.mlm - ld.mlm + 1):Ltmp3] <-
c(tmp7a, tmp7b, tmp7c)
par1or2 <- 1 # 2
tmp3.TF <- c(TRUE, la.mix > 0, la.mix > 1,
li.mix > 0, li.mix > 1,
ld.mix > 0, ld.mix > 1,
la.mlm > 0, li.mlm > 0, ld.mlm > 0)
indeta.finish <- cumsum(c(par1or2, 1, par1or2,
1, par1or2,
1, par1or2,
la.mlm, li.mlm, ld.mlm,
ld.mlm + 1) * c(tmp3.TF, 1))
indeta.launch <- c(1, 1 + head(indeta.finish, -1))
indeta.launch <- head(indeta.launch, -1)
indeta.finish <- head(indeta.finish, -1)
indeta.launch[!tmp3.TF] <- NA # Not to be accessed
indeta.finish[!tmp3.TF] <- NA # Not to be accessed
indeta <- cbind(launch = indeta.launch,
finish = indeta.finish)
rownames(indeta) <- c("shape.p",
"pobs.mix", "shape.a",
"pstr.mix", "shape.i",
"pdip.mix", "shape.d",
"pobs.mlm", "pstr.mlm", "pdip.mlm")
M1 <- max(indeta, na.rm = TRUE)
predictors.names <- tmp3 # Passed into @infos and @initialize.
blurb1 <- "L" # zz1
if (la.mlm + la.mix) blurb1 <- "Generally-altered l"
if (li.mlm + li.mix) blurb1 <- "Generally-inflated l"
if (ltrunc.use) blurb1 <- "Generally-truncated l"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && !ltrunc.use)
blurb1 <- "Generally-altered and -inflated l"
if ( (la.mlm + la.mix) && !(li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered and -truncated l"
if (!(la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-inflated and -truncated l"
if ( (la.mlm + la.mix) && (li.mlm + li.mix) && ltrunc.use)
blurb1 <- "Generally-altered, -inflated and -truncated l"
if (ld.mlm + ld.mix) blurb1 <-
c(blurb1,
if (la.mlm + la.mix + li.mlm + li.mix) "and " else "Generally",
"-deflated ")
new("vglmff",
blurb = c(blurb1, "ogarithmic regression\n",
"(GAITD-Log(shape.p)-",
"Log(shape.a)-MLM-",
"Log(shape.i)-MLM-\n",
"Log(shape.d)-MLM generally)\n\n",
"Links: ",
namesof("shape.p", lshape.p, earg = eshape.p,
tag = FALSE),
if (la.mix > 0) c(", ", "multilogit(pobs.mix)"),
if (la.mix > 1) c(", ",
namesof("shape.a", lshape.a, eshape.a, tag = FALSE)),
if (la.mix && li.mix) ", \n ",
if (li.mix > 0) c( if (la.mix) "" else ", ",
"multilogit(pstr.mix)"),
if (li.mix > 1) c(", ",
namesof("shape.i", lshape.i, eshape.i, tag = FALSE)),
if (li.mix && ld.mix) ", \n ",
if (ld.mix > 0) c( if (li.mix) "" else ", ",
"multilogit(pdip.mix)"),
if (ld.mix > 1) c(", ",
namesof("shape.d", lshape.d, eshape.d, tag = FALSE)),
if (la.mlm) paste0(",\n",
paste0(" multilogit(", tmp7a, collapse = "),\n"),
")") else NULL,
if (li.mlm) paste0(",\n",
paste0(" multilogit(", tmp7b, collapse = "),\n"),
")") else NULL,
if (ld.mlm) paste0(",\n",
paste0(" multilogit(", tmp7c, collapse = "),\n"),
")") else NULL),
constraints = eval(substitute(expression({
M1 <- max(extra$indeta, na.rm = TRUE)
la.mix <- ( .la.mix )
li.mix <- ( .li.mix )
ld.mix <- ( .ld.mix )
la.mlm <- ( .la.mlm )
li.mlm <- ( .li.mlm )
ld.mlm <- ( .ld.mlm )
use.mat.mlm.a <- if (la.mlm) {
if ( .parallel.a ) matrix(1, la.mlm, 1) else diag(la.mlm)
} else {
NULL
}
use.mat.mlm.i <- if (li.mlm) {
if ( .parallel.i ) matrix(1, li.mlm, 1) else diag(li.mlm)
} else {
NULL
}
use.mat.mlm.d <- if (ld.mlm) {
if ( .parallel.d ) matrix(1, ld.mlm, 1) else diag(ld.mlm)
} else {
NULL
}
if (la.mlm + li.mlm + ld.mlm == 0) {
Use.mat <- use.mat.mlm <- cbind(M) # shape.p only
}
if (la.mlm + li.mlm + ld.mlm) {
nc1 <- if (length(use.mat.mlm.a)) ncol(use.mat.mlm.a) else 0
nc2 <- if (length(use.mat.mlm.i)) ncol(use.mat.mlm.i) else 0
nc3 <- if (length(use.mat.mlm.d)) ncol(use.mat.mlm.d) else 0
use.mat.mlm <- cbind(1, matrix(0, 1, nc1 + nc2 + nc3))
if (la.mlm)
use.mat.mlm <- rbind(use.mat.mlm,
cbind(matrix(0, la.mlm, 1),
use.mat.mlm.a,
if (length(use.mat.mlm.i) == 0)
NULL else matrix(0, la.mlm, nc2),
if (length(use.mat.mlm.d) == 0)
NULL else matrix(0, la.mlm, nc3)))
if (li.mlm )
use.mat.mlm <-
rbind(use.mat.mlm,
cbind(matrix(0, li.mlm, 1 + nc1),
use.mat.mlm.i,
matrix(0, li.mlm, nc3)))
if (ld.mlm)
use.mat.mlm <-
rbind(use.mat.mlm, # zz1 next line:
cbind(matrix(0, ld.mlm, 1 + nc1 + nc2),
use.mat.mlm.d))
} # la.mlm + li.mlm
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
use.mat.mix <- cm3gaitd( .eq.ap , .eq.ip , .eq.dp , npar = 1)
tmp3.subset <- tmp3.TF[-(8:10)]
use.mat.mix <- use.mat.mix[tmp3.subset, , drop = FALSE]
notall0 <- function(x) !all(x == 0)
use.mat.mix <- use.mat.mix[, apply(use.mat.mix, 2, notall0),
drop = FALSE]
if (la.mix + li.mix + ld.mix > 0)
Use.mat <- use.mat.mix
if (la.mlm + li.mlm + ld.mlm > 0) {
Use.mat <- rbind(use.mat.mix,
matrix(0, nrow(use.mat.mlm) - 1, # bottom
ncol(use.mat.mix)))
Use.mat <- cbind(Use.mat,
matrix(0, nrow(Use.mat), # RHS
ncol(use.mat.mlm) - 1))
Use.mat[row(Use.mat) > nrow(use.mat.mix) &
col(Use.mat) > ncol(use.mat.mix)] <- use.mat.mlm[-1, -1]
} # la.mlm + li.mlm + ld.mlm > 0
if (is.null(constraints)) {
constraints <-
cm.VGAM(Use.mat, x = x, apply.int = TRUE, # FALSE
bool = .eq.ap || .eq.ip || .eq.dp ||
.parallel.a || .parallel.i || .parallel.d ,
constraints = constraints) # FALSE
} # is.null(constraints)
if (la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm)
constraints <-
cm.zero.VGAM(constraints, x = x, .zero , M = M, M1 = M1,
predictors.names = paste0(predictors.names,
names(predictors.names)))
}),
list( .zero = zero, .tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.eq.ap = eq.ap, .eq.ip = eq.ip, .eq.dp = eq.dp,
.parallel.a = parallel.a, .parallel.i = parallel.i,
.parallel.d = parallel.d,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix ))),
infos = eval(substitute(function(...) {
list(M1 = .M1 ,
Q1 = 1,
dpqrfun = "gaitdlog",
link = .predictors.names , # ...strips... from above
link1parameter = as.logical( .lall.len <= 2), # <= 1 safer
mixture.links = any(c( .la.mlm , .li.mlm , .ld.mlm ,
.la.mix , .li.mix ,
.ld.mix ) > 1), # FALSE if NULL
a.mix = as.vector( .a.mix ), # Handles NULL
a.mlm = as.vector( .a.mlm ),
i.mix = as.vector( .i.mix ),
i.mlm = as.vector( .i.mlm ),
d.mix = as.vector( .d.mix ),
d.mlm = as.vector( .d.mlm ),
truncate = as.vector( .truncate ),
max.support = as.vector( .max.support ),
Support = c( .lowsup , Inf, 1), # a(b)c format as a,c,b.
expected = TRUE,
multipleResponses = FALSE, # poissonff can b called ifTRUE
parameters.names = names( .predictors.names ),
parent.name = c("logff", "log"),
type.fitted = as.vector( .type.fitted ),
type.fitted.choices = ( .type.fitted.choices ),
baseparams.argnames = "shape",
MM1 = 1, # One parameter for 1 response (shape). Needed.
zero = .zero )
}, list( .zero = zero, .lowsup = lowsup,
.type.fitted = type.fitted,
.type.fitted.choices = type.fitted.choices,
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.la.mlm = la.mlm, .li.mlm = li.mlm, .ld.mlm = ld.mlm,
.la.mix = la.mix, .li.mix = li.mix, .ld.mix = ld.mix,
.truncate = truncate, .max.support = max.support,
.predictors.names = predictors.names,
.M1 = M1, .lall.len = lall.len
))),
rqresslot = eval(substitute(
function(mu, y, w, eta, extra = NULL) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least 1 shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vecs
ind.shape.z <- c(na.omit(ind.shape.z)) # At least 1 value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
scrambleseed <- runif(1) # To scramble the seed
qnorm(runif(length(y),
pgaitdlog(y - 1, shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm),
pgaitdlog(y , shape.p = shape.p,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)))
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
initialize = eval(substitute(expression({
extra$indeta <- ( .indeta ) # Avoids recomputing it
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix +
la.mlm + li.mlm + ld.mlm
truncate <- as.vector( .truncate )
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(y)
M <- NOS * M1
tmp3.TF <- ( .tmp3.TF )
temp5 <-
w.y.check(w = w, y = y,
Is.nonnegative.y = TRUE,
Is.integer.y = TRUE,
ncol.w.max = 1, # Since max.support = 9 is possible
ncol.y.max = 1,
out.wy = TRUE, colsyperw = 1, maximize = TRUE)
w <- temp5$w
y <- temp5$y
glist <- y.gaitcombo.check(y, truncate = truncate,
a.mlm = a.mlm, a.mix = a.mix,
i.mlm = i.mlm, i.mix = i.mix,
d.mlm = d.mlm, d.mix = d.mix,
max.support = .max.support ,
min.support = .min.support )
extra$skip.mix.a <- glist$skip.mix.a
extra$skip.mix.i <- glist$skip.mix.i
extra$skip.mix.d <- glist$skip.mix.d
extra$skip.mlm.a <- glist$skip.mlm.a
extra$skip.mlm.i <- glist$skip.mlm.i
extra$skip.mlm.d <- glist$skip.mlm.d
extra$NOS <- NOS <- ncoly <- ncol(y) # Number of species
extra$type.fitted <- as.vector( .type.fitted )
extra$mux.init <- as.vector( .mux.init )
extra$colnames.y <- colnames(y)
extra$M1 <- M1
extra$index.M <- iam(NA, NA, M, both = TRUE) # Used in @weight
predictors.names <- ( .predictors.names ) # Got it, named
if (!length(etastart)) {
shape.p.init <- if (length( .ishape.p )) .ishape.p else {
logff.Loglikfun <- function(shapeval, y, x, w, extraargs) {
sum(c(w) * dlog(x = y, shape = shapeval, log = TRUE))
}
shape.p.grid <- ( .gshape.p )
grid.search(shape.p.grid,
objfun = logff.Loglikfun, y = y, w = w)
}
shape.p.init <- rep(shape.p.init, length = n)
shape.d.init <-
shape.a.init <- shape.i.init <- shape.p.init # Needed
etastart <- matrix(nrow = n, ncol = M,
theta2eta(shape.p.init, .lshape.p , earg = .eshape.p ))
mux.more.a <- extra$mux.init[1] # 0.75 Err 2 bit smaller
init.pobs.mix <- numeric(n)
if (tmp3.TF[ 2]) { # la.mix > 0
init.pobs.mix <- if (length( .ipobs.mix )) {
rep_len( .ipobs.mix , n)
} else {
is.a.mix1 <- rowSums(extra$skip.mix.a) > 0
rep(mux.more.a * sum(w[is.a.mix1]) / sum(w), n)
}
} # la.mix > 0
if (tmp3.TF[ 3]) { # Assign coln 3; la.mix > 1
shape.a.init <- if (length( .ishape.a ))
rep_len( .ishape.a , n) else shape.p.init # A vector
etastart[, 3] <-
theta2eta(shape.a.init, .lshape.a , earg = .eshape.a )
}
init.pstr.mix <- init.pdip.mix <- numeric(n)
try.gridsearch.pstr.mix <- FALSE
if (tmp3.TF[ 4]) { # li.mix > 0
init.pstr.mix <- if (length( .ipstr.mix )) {
rep_len( .ipstr.mix , n)
} else {
try.gridsearch.pstr.mix <- TRUE
numeric(n) # Overwritten by gridsearch
}
} # li.mix > 0
if (tmp3.TF[ 5]) { # li.mix > 1
shape.i.init <- if (length( .ishape.i ))
rep_len( .ishape.i , n) else shape.p.init # A vector
etastart[, (extra$indeta[5, 'launch'])] <-
theta2eta(shape.i.init, .lshape.i , earg = .eshape.i )
} # li.mix > 1
if (tmp3.TF[ 8]) { # la.mlm
init.pobs.mlm <- if (length( .ipobs.mlm )) {
matrix( .ipobs.mlm , n, la.mlm, byrow = .byrow.aid )
} else {
mux.more.a <- extra$mux.init[1]
init.pobs.mlm <- colSums(c(w) *
extra$skip.mlm.a) / colSums(w)
init.pobs.mlm <- init.pobs.mlm * as.vector( mux.more.a )
matrix(init.pobs.mlm, n, la.mlm, byrow = TRUE)
}
} else {
init.pobs.mlm <- matrix(0, n, 1)
}
try.gridsearch.pstr.mlm <- FALSE
if (tmp3.TF[ 9]) { # li.mlm
try.gridsearch.pstr.mlm <- !(length( .ipstr.mlm ))
init.pstr.mlm <- 0 # Might be overwritten by gridsearch
if (length( .ipstr.mlm ))
init.pstr.mlm <- as.vector( .ipstr.mlm )
init.pstr.mlm <- matrix(init.pstr.mlm, n, li.mlm,
byrow = .byrow.aid )
} else {
init.pstr.mlm <- matrix(0, n, 1)
}
init.pdip.mlm <- matrix(0, n, 2) # rowSums() needs > 1 colns.
gaitdlog.Loglikfun1.mix <-
function(pstr.mix.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mix = pstr.mix.val,
pstr.mlm = extraargs$pstr.mlm, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdlog.Loglikfun1.mlm <-
function(pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mlm = pstr.mlm.val,
pstr.mix = extraargs$pstr.mix, # Differs here
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
gaitdlog.Loglikfun2 <-
function(pstr.mix.val, pstr.mlm.val, y, x, w, extraargs) {
sum(c(w) *
dgaitdlog(y, pstr.mix = pstr.mix.val,
pstr.mlm = pstr.mlm.val,
shape.p = extraargs$shape.p,
shape.a = extraargs$shape.a,
shape.i = extraargs$shape.i,
shape.d = extraargs$shape.d,
a.mix = extraargs$a.mix,
a.mlm = extraargs$a.mlm,
i.mix = extraargs$i.mix,
i.mlm = extraargs$i.mlm,
d.mix = extraargs$d.mix,
d.mlm = extraargs$d.mlm,
max.support = extraargs$max.support,
truncate = extraargs$truncate,
pobs.mix = extraargs$pobs.mix,
pobs.mlm = extraargs$pobs.mlm,
pdip.mix = extraargs$pdip.mix,
pdip.mlm = extraargs$pdip.mlm, log = TRUE))
}
if (li.mix + li.mlm) {
extraargs <- list(
shape.p = shape.p.init,
shape.a = shape.a.init,
shape.i = shape.i.init,
shape.d = shape.d.init,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
pobs.mix = init.pobs.mix ,
pobs.mlm = init.pobs.mlm ,
pdip.mix = init.pdip.mix ,
pdip.mlm = init.pdip.mlm )
pre.warn <- options()$warn
options(warn = -1) # Ignore warnings during gridsearch
try.this <-
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
grid.search2( .gpstr.mix , .gpstr.mlm ,
objfun = gaitdlog.Loglikfun2,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mix) {
extraargs$pstr.mlm <- init.pstr.mlm
grid.search ( .gpstr.mix ,
objfun = gaitdlog.Loglikfun1.mix,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
} else if (try.gridsearch.pstr.mlm) {
extraargs$pstr.mix <- init.pstr.mix
grid.search ( .gpstr.mlm ,
objfun = gaitdlog.Loglikfun1.mlm,
y = y, w = w, extraargs = extraargs,
ret.objfun = TRUE)
}
options(warn = pre.warn) # Restore warnings
if (any(is.na(try.this)))
warning("gridsearch returned NAs. It's going to crash.",
immediate. = TRUE)
if (try.gridsearch.pstr.mix && try.gridsearch.pstr.mlm) {
init.pstr.mix <- rep_len(try.this["Value1"], n)
init.pstr.mlm <- matrix(try.this["Value2"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'",
" and/or 'gpstr.mlm = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mix) {
init.pstr.mix <- rep_len(try.this["Value"], n)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mix = seq(5) / 100'.")
} else if (try.gridsearch.pstr.mlm) {
init.pstr.mlm <- matrix(try.this["Value"], n, li.mlm)
if (any(is.na(try.this)))
stop("Crashing. ",
"Try something like 'gpstr.mlm = seq(5) / 100'.")
}
} # la.mix + lnf.mix
mux.more.d <- extra$mux.init[3]
if (ld.mix) {
init.pdip.mix <- if (length( .ipdip.mix ))
rep_len( .ipdip.mix, n) else {
is.d.mix1 <- rowSums(extra$skip.mix.d) > 0
rep(mux.more.d * sum(w[is.d.mix1]) / sum(w), n)
}
} # ld.mix
if (ld.mlm) {
init.pdip.mlm <- if (length( .ipdip.mlm ))
matrix( .ipdip.mlm, n, ld.mlm, byrow = TRUE) else {
is.d.mlm1 <- rowSums(extra$skip.mlm.d) > 0
matrix(mux.more.d * (sum(w[is.d.mlm1]) / sum(w)) / ld.mlm,
n, ld.mlm)
}
} # ld.mlm
while (any((vecTF <- init.pobs.mix + init.pstr.mix + # -
init.pdip.mix +
rowSums(init.pobs.mlm) +
rowSums(init.pstr.mlm) + # -
rowSums(init.pdip.mlm) > 0.96875))) {
init.pobs.mix[vecTF] <- 0.875 * init.pobs.mix[vecTF]
init.pstr.mix[vecTF] <- 0.875 * init.pstr.mix[vecTF]
init.pdip.mix[vecTF] <- 0.875 * init.pdip.mix[vecTF]
init.pobs.mlm[vecTF, ] <- 0.875 * init.pobs.mlm[vecTF, ]
init.pstr.mlm[vecTF, ] <- 0.875 * init.pstr.mlm[vecTF, ]
init.pdip.mlm[vecTF, ] <- 0.875 * init.pdip.mlm[vecTF, ]
} # while
Numer.init1 <- 1 - rowSums(init.pobs.mlm) -
rowSums(init.pstr.mlm) - # +
rowSums(init.pdip.mlm) -
init.pobs.mix - init.pstr.mix - # +
init.pdip.mix # Differs from 'Numer'.
etastart.z <- if (lall.len == 0) NULL else {
tmp.mat <- cbind(if (tmp3.TF[ 2]) init.pobs.mix else NULL,
if (tmp3.TF[ 4]) init.pstr.mix else NULL,
if (tmp3.TF[ 6]) init.pdip.mix else NULL,
if (tmp3.TF[ 8]) init.pobs.mlm else NULL,
if (tmp3.TF[ 9]) init.pstr.mlm else NULL,
if (tmp3.TF[10]) init.pdip.mlm else NULL,
Numer.init1)
multilogitlink(tmp.mat)
} # etastart.z
if (!is.matrix(etastart.z)) etastart.z <- cbind(etastart.z)
nextone <- 1 # Might not be used actually
if (tmp3.TF[ 2]) {
etastart[, 2] <- etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 4]) { # Coln 2 or 4
etastart[, (extra$indeta[4, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 6]) { # Coln 2 or 4 or 6
etastart[, (extra$indeta[6, 'launch'])] <-
etastart.z[, nextone]
nextone <- nextone + 1
}
if (tmp3.TF[ 8]) {
ind8 <- (extra$indeta[8, 'launch']):(
extra$indeta[8, 'finish'])
etastart[, ind8] <- etastart.z[, nextone:(nextone+
la.mlm - 1)]
nextone <- nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (extra$indeta[9, 'launch']):(
extra$indeta[9, 'finish'])
etastart[, ind9] <- etastart.z[, nextone:(nextone+
li.mlm - 1)]
nextone <- nextone + li.mlm
}
if (tmp3.TF[10]) {
ind0 <- (extra$indeta[10, 'launch']):(
extra$indeta[10, 'finish'])
etastart[, ind0] <- etastart.z[, nextone:(nextone +
ld.mlm - 1)]
if (ncol(etastart.z) != nextone + ld.mlm - 1)
stop("miscalculation")
}
}
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.ishape.p = ishape.p,
.ishape.a = ishape.a,
.ishape.i = ishape.i,
.ishape.d = ishape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.lpdip.mix = lpdip.mix,
.epdip.mix = epdip.mix,
.ipstr.mix = ipstr.mix, .ipobs.mix = ipobs.mix,
.ipstr.mlm = ipstr.mlm, .ipobs.mlm = ipobs.mlm,
.ipdip.mix = ipdip.mix,
.ipdip.mlm = ipdip.mlm,
.byrow.aid = byrow.aid,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support,
.min.support = lowsup,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.predictors.names = predictors.names,
.mux.init = mux.init,
.gshape.p = gshape.p,
.gpstr.mix = gpstr.mix, # .gpdip.mix = gpdip.mix,
.gpstr.mlm = gpstr.mlm, # .gpdip.mlm = gpdip.mlm,
.ishrinkage = ishrinkage, .probs.y = probs.y,
.indeta = indeta,
.imethod = imethod, .type.fitted = type.fitted ))),
linkinv = eval(substitute(function(eta, extra = NULL) {
n.obs <- NROW(eta)
type.fitted <-
if (length(extra$type.fitted)) extra$type.fitted else {
warning("cannot find 'type.fitted'. Returning the 'mean'.")
"mean"
}
type.fitted <-
match.arg(type.fitted[1],
c("mean", "shapes",
"pobs.mlm", "pstr.mlm", "pdip.mlm",
"pobs.mix", "pstr.mix", "pdip.mix",
"Pobs.mix", "Pstr.mix", "Pdip.mix",
"nonspecial", "Numer", "Denom.p",
"sum.mlm.i", "sum.mix.i",
"sum.mlm.d", "sum.mix.d",
"ptrunc.p", "cdf.max.s"))[1]
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
morework <- type.fitted != "mean" # For efficiency
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF ) # Logical of length 10.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- NCOL(eta) / M1
Bits <- moments.gaitdcombo.log(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
if (morework) {
Denom.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
if (any(Denom.p == 0)) {
smallval <- min(Denom.p[Denom.p > 0])
Denom.p[Denom.p == 0] <- 1e-09 # smallval
warning("0s found in variable 'Denom.p'. Trying to fix it.")
}
Numer <- c(1 - pobs.mix - pstr.mix + pdip.mix -
(if (la.mlm) rowSums(pobs.mlm) else 0) -
(if (li.mlm) rowSums(pstr.mlm) else 0) +
(if (ld.mlm) rowSums(pdip.mlm) else 0))
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom.p)
} # morework
if (!la.mlm && type.fitted %in% c("pobs.mlm")) {
warning("No altered MLM values; returning an NA")
return(NA)
}
if (!li.mlm && type.fitted %in% c("sum.mlm.i", "pstr.mlm")) {
warning("No inflated MLM values; returning an NA")
return(NA)
}
if (!ld.mlm && type.fitted %in% c("sum.mlm.d", "pdip.mlm")) {
warning("No deflated MLM values; returning an NA")
return(NA)
}
if (!la.mix && type.fitted %in% c("Pobs.mix")) {
warning("No altered mixture values; returning an NA")
return(NA)
}
if (!li.mix && type.fitted %in% c("sum.mix.i", "Pstr.mix")) {
warning("No inflated mixture values; returning an NA")
return(NA)
}
if (!ld.mix && type.fitted %in% c("sum.mix.d", "Pdip.mix")) {
warning("No deflated mixture values; returning an NA")
return(NA)
}
if (la.mix && morework) {
tmp13 <- # dpois() does not retain the matrix format
dlog(matrix(a.mix, n.obs, la.mix, byrow = TRUE),
matrix(shape.a, n.obs, la.mix)) / (
c(Bits[["SumA0.mix.a"]]))
dim(tmp13) <- c(n.obs, la.mix)
dimnames(tmp13) <- list(rownames(eta),
as.character(a.mix))
propn.mat.a <- tmp13
} # la.mix
if (li.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dlog(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.i, n.obs, li.mix)) / (
c(Bits[["SumI0.mix.i"]]))
dim(tmp55) <- c(n.obs, li.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(i.mix))
propn.mat.i <- tmp55 # Correct dimension
} # li.mix
if (ld.mix && morework) {
tmp55 <- # dpois() does not retain the matrix format
dlog(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.d, n.obs, ld.mix)) / (
c(Bits[["SumD0.mix.d"]]))
dim(tmp55) <- c(n.obs, ld.mix)
dimnames(tmp55) <- list(rownames(eta),
as.character(d.mix))
propn.mat.d <- tmp55 # Correct dimension
} # ld.mix
ans <- switch(type.fitted,
"mean" = Bits[["mean"]], # Unconditional mean
"shapes" = cbind(shape.p,
if (tmp3.TF[ 3]) shape.a else NULL,
if (tmp3.TF[ 5]) shape.i else NULL,
if (tmp3.TF[ 7]) shape.d else NULL),
"pobs.mlm" = pobs.mlm, # aka omegamat, n x la.mlm
"pstr.mlm" = pstr.mlm, # aka phimat, n x li.mlm
"pdip.mlm" = pdip.mlm, # aka psimat, n x ld.mlm
"pobs.mix" = pobs.mix, # n-vector
"pstr.mix" = pstr.mix, # n-vector
"pdip.mix" = pdip.mix, # n-vector
"Pobs.mix" = c(pobs.mix) * propn.mat.a, # matrix
"Pstr.mix" = c(pstr.mix) * propn.mat.i,
"Pdip.mix" = c(pdip.mix) * propn.mat.d,
"nonspecial" = probns,
"Numer" = Numer,
"Denom.p" = Denom.p,
"sum.mlm.i" = pstr.mlm + Numer *
dlog(matrix(i.mlm, n.obs, li.mlm, byrow = TRUE),
matrix(shape.p, n.obs, li.mlm)) / Denom.p,
"sum.mlm.d" = -pdip.mlm + Numer *
dlog(matrix(d.mlm, n.obs, ld.mlm, byrow = TRUE),
matrix(shape.p, n.obs, ld.mlm)) / Denom.p,
"sum.mix.i" = c(pstr.mix) * propn.mat.i + Numer *
dlog(matrix(i.mix, n.obs, li.mix, byrow = TRUE),
matrix(shape.p, n.obs, li.mix)) / Denom.p,
"sum.mix.d" = -c(pdip.mix) * propn.mat.d + Numer *
dlog(matrix(d.mix, n.obs, ld.mix, byrow = TRUE),
matrix(shape.p, n.obs, ld.mix)) / Denom.p,
"ptrunc.p" = Bits[["SumT0.p"]] + 1 - Bits[["cdf.max.s"]],
"cdf.max.s" = Bits[["cdf.max.s"]]) # Pr(y <= max.support)
ynames.pobs.mlm <- as.character(a.mlm) # Works with NULLs
ynames.pstr.mlm <- as.character(i.mlm) # Works with NULLs
ynames.pdip.mlm <- as.character(d.mlm) # Works with NULLs
if (length(ans))
label.cols.y(ans, NOS = NOS, colnames.y =
switch(type.fitted,
"shapes" = c("shape.p", "shape.a", # Some colns NA
"shape.i", "shape.d")[(tmp3.TF[c(1, 3, 5, 7)])],
"Pobs.mix" = as.character(a.mix),
"sum.mix.i" = , #
"Pstr.mix" = as.character(i.mix),
"sum.mix.d" = , #
"Pdip.mix" = as.character(d.mix),
"pobs.mlm" = ynames.pobs.mlm,
"sum.mlm.i" = , #
"pstr.mlm" = ynames.pstr.mlm,
"sum.mlm.d" = , #
"pdip.mlm" = ynames.pdip.mlm,
extra$colnames.y)) else ans
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
last = eval(substitute(expression({
pred.names <- c( .predictors.names ) # Save it
link.names <- as.vector( .predictors.names )
parameter.names <- names(pred.names)
predictors.names <- NULL
for (jay in seq(M))
predictors.names <- c(predictors.names,
namesof(parameter.names[jay], link.names[jay], tag = FALSE,
earg = list())) # This isnt perfect; info is lost
misc$predictors.names <- predictors.names # Useful for coef()
misc$link <- link.names #
names(misc$link) <- parameter.names #
misc$earg <- vector("list", M1)
names(misc$earg) <- names(misc$link)
misc$earg[[1]] <- ( .eshape.p ) # First one always there
iptr <- 1
if (tmp3.TF[ 2])
misc$earg[[(iptr <- iptr + 1)]] <- list() # multilogitlink
if (tmp3.TF[ 3])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.a )
if (tmp3.TF[ 4])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 5])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.i )
if (tmp3.TF[ 6])
misc$earg[[(iptr <- iptr + 1)]] <- list() # See below
if (tmp3.TF[ 7])
misc$earg[[(iptr <- iptr + 1)]] <- ( .eshape.d )
if (tmp3.TF[ 8]) { # la.mlm
for (ii in seq(la.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # la.mlm
if (tmp3.TF[ 9]) { # li.mlm
for (ii in seq(li.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # li.mlm
if (tmp3.TF[10]) { # ld.mlm
for (ii in seq(ld.mlm)) {
misc$earg[[(iptr <- iptr + 1)]] <- list()
} # ii
} # ld.mlm
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.predictors.names = predictors.names,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
loglikelihood = eval(substitute(
function(mu, y, w, residuals = FALSE, eta,
extra = NULL, summation = TRUE) {
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed and doesnt corrupt the answer
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # An MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1)
warning("fitted probabilities numerically 0 or 1 occurred")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
if (residuals) {
stop("loglikelihood residuals not implemented yet")
} else {
ll.elts <- c(w) *
dgaitdlog(y, shape.p, log = TRUE, # byrow.aid = F,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = truncate,
max.support = as.vector( .max.support ),
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm)
if (summation) {
sum(ll.elts)
} else {
ll.elts
}
}
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
vfamily = c("gaitdlog"),
validparams = eval(substitute(function(eta, y, extra = NULL) {
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
small. <- 1e-14
pobs.mix <- pstr.mix <- pdip.mix <- small. # 4 rowSums():
pobs.mlm <- pstr.mlm <- pdip.mlm <- matrix(small., NROW(eta), 1)
shape.a <- shape.i <- shape.d <- 0.5 # Needed
if (!is.matrix(eta)) eta <- as.matrix(eta)
shape.p <-
cbind(eta2theta(eta[, 1], .lshape.p , earg = .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 1] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
okay.mlm <-
all(is.finite(pobs.mlm)) && all(0 < pobs.mlm) &&
all(is.finite(pstr.mlm)) && all(0 < pstr.mlm) &&
all(is.finite(pdip.mlm)) && all(0 < pdip.mlm)
okay.mix <-
all(is.finite(shape.p)) && all(0 < shape.p) && all(shape.p < 1) &&
all(is.finite(shape.a)) && all(0 < shape.a) && all(shape.a < 1) &&
all(is.finite(shape.i)) && all(0 < shape.i) && all(shape.i < 1) &&
all(is.finite(shape.d)) && all(0 < shape.d) && all(shape.d < 1) &&
all(is.finite(pobs.mix)) && all(0 < pobs.mix) &&
all(is.finite(pstr.mix)) && all(0 < pstr.mix) &&
all(is.finite(pdip.mix)) && all(0 < pdip.mix) &&
all(pobs.mix + pstr.mix + pdip.mix +
rowSums(pobs.mlm) + rowSums(pstr.mlm) +
rowSums(pdip.mlm) < 1) # Combined
okay.mlm && okay.mix
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
simslot = eval(substitute(
function(object, nsim) {
pwts <- if (length(pwts <- object@prior.weights) > 0)
pwts else weights(object, type = "prior")
if (any(pwts != 1))
warning("ignoring prior weights")
eta <- predict(object)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
extra <- object@extra
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
tmp3.TF <- ( .tmp3.TF )
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A AMLM was fitted
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
inverse = TRUE) # rowSums == 1
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
rgaitdlog(nsim * length(shape.p), shape.p,
pobs.mlm = pobs.mlm, pstr.mlm = pstr.mlm,
pobs.mix = pobs.mix, pstr.mix = pstr.mix,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
truncate = .truncate , max.support = .max.support )
}, list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.tmp3.TF = tmp3.TF,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.truncate = truncate, .max.support = max.support ))),
deriv = eval(substitute(expression({
tmp3.TF <- ( .tmp3.TF )
calA.p <- tmp3.TF[ 2]
calI.p <- tmp3.TF[ 4]
calD.p <- tmp3.TF[ 6]
calA.np <- tmp3.TF[ 8]
calI.np <- tmp3.TF[ 9]
calD.np <- tmp3.TF[10]
Denom1.a <- Denom1.i <- Denom1.d <-
Denom2.i <- Denom2.d <- 0 # Denom2.a is unneeded
if (!is.matrix(eta)) eta <- as.matrix(eta)
la.mix <- length((a.mix <- as.vector( .a.mix )))
li.mix <- length((i.mix <- as.vector( .i.mix )))
ld.mix <- length((d.mix <- as.vector( .d.mix )))
la.mlm <- length((a.mlm <- as.vector( .a.mlm )))
li.mlm <- length((i.mlm <- as.vector( .i.mlm )))
ld.mlm <- length((d.mlm <- as.vector( .d.mlm )))
truncate <- as.vector( .truncate )
max.support <- as.vector( .max.support )
lall.len <- la.mix + li.mix + ld.mix + la.mlm + li.mlm + ld.mlm
pobs.mix <- pstr.mix <- pdip.mix <- 0 # 4 rowSums()
pobs.mlm <- pstr.mlm <- pdip.mlm <- 0 # matrix(0, NROW(eta), 1)
shape.p <- cbind(eta2theta(eta[, 1], .lshape.p , .eshape.p ))
ind.shape.z <- 1 # Points to shape.p only.
shape.a <- shape.i <-
shape.d <- shape.p # Needed; and answer not corrupted
if (any(tmp3.TF[c(3, 5, 7)])) { # At least one shape.[aid]
ind.shape.z <- extra$indeta[c(1, 3, 5, 7), 'launch'] # Vectors
ind.shape.z <- c(na.omit(ind.shape.z)) # At least one value
shape.a <- if (!tmp3.TF[ 3]) shape.p else
eta2theta(eta[, extra$indeta[3, 1]], .lshape.a , .eshape.a )
shape.i <- if (!tmp3.TF[ 5]) shape.p else
eta2theta(eta[, extra$indeta[5, 1]], .lshape.i , .eshape.i )
shape.d <- if (!tmp3.TF[ 7]) shape.p else
eta2theta(eta[, extra$indeta[7, 1]], .lshape.d , .eshape.d )
} # la.mix + li.mix + ld.mix > 0
if (lall.len) { # A MLM was fitted.
allprobs <-
multilogitlink(eta[, -ind.shape.z, drop = FALSE],
refLevel = "(Last)", # Make sure
inverse = TRUE) # rowSums == 1
minprob.baseline <- min(allprobs[, ncol(allprobs)], na.rm = TRUE)
if (anyNA(allprobs))
warning("there are NAs here in slot linkinv")
if (min(allprobs) == 0 || max(allprobs) == 1) {
warning("fitted probabilities numerically 0 or 1 occurred")
} else
if (minprob.baseline < 0.10)
warning("Minimum baseline (reserve) probability close to 0")
if (control$trace)
cat("Minimum baseline (reserve) probability = ",
format(minprob.baseline, digits = 3), "\n")
Nextone <- 0 # Might not be used actually; 0, not 1
if (tmp3.TF[ 2])
pobs.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 4])
pstr.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 6])
pdip.mix <- allprobs[, (Nextone <- Nextone + 1)]
if (tmp3.TF[ 8]) {
ind8 <- (Nextone + 1):(Nextone + la.mlm)
pobs.mlm <- allprobs[, ind8, drop = FALSE]
dimnames(pobs.mlm) <- list(rownames(eta),
as.character(a.mlm))
Nextone <- Nextone + la.mlm
}
if (tmp3.TF[ 9]) {
ind9 <- (Nextone + 1):(Nextone + li.mlm)
pstr.mlm <- allprobs[, ind9, drop = FALSE]
dimnames(pstr.mlm) <- list(rownames(eta),
as.character(i.mlm))
Nextone <- Nextone + li.mlm
}
if (tmp3.TF[10]) {
ind10 <- (Nextone + 1):(Nextone + ld.mlm)
pdip.mlm <- allprobs[, ind10, drop = FALSE]
dimnames(pdip.mlm) <- list(rownames(eta),
as.character(d.mlm))
Nextone <- Nextone + ld.mlm # Not needed
}
} # lall.len
ltruncat <- length(truncate)
M1 <- max(extra$indeta, na.rm = TRUE)
NOS <- ncol(eta) / M1 # extra$NOS
if (NOS != 1) stop("can only handle 1 response")
is.a.mixed <- if (tmp3.TF[ 2])
rowSums(extra$skip.mix.a) > 0 else rep(FALSE, n)
is.i.mixed <- if (tmp3.TF[ 4])
rowSums(extra$skip.mix.i) > 0 else rep(FALSE, n)
is.d.mixed <- if (tmp3.TF[ 6])
rowSums(extra$skip.mix.d) > 0 else rep(FALSE, n)
is.a.mlmed <- if (tmp3.TF[ 8])
rowSums(extra$skip.mlm.a) > 0 else rep(FALSE, n)
is.i.mlmed <- if (tmp3.TF[ 9])
rowSums(extra$skip.mlm.i) > 0 else rep(FALSE, n)
is.d.mlmed <- if (tmp3.TF[10])
rowSums(extra$skip.mlm.d) > 0 else rep(FALSE, n)
is.ns <- !is.a.mlmed & !is.i.mlmed & !is.d.mlmed &
!is.a.mixed & !is.i.mixed & !is.d.mixed # & !is.truncd
A8.p <- -1 / log1p(-shape.p)
A8.a <- -1 / log1p(-shape.a)
A8.i <- -1 / log1p(-shape.i)
prob.mlm.a <- if (la.mlm) rowSums(pobs.mlm) else 0 # scalar okay
prob.mlm.i <- if (li.mlm) rowSums(pstr.mlm) else 0 # scalar okay
prob.mlm.d <- if (ld.mlm) rowSums(pdip.mlm) else 0 # scalar okay
pmf.deriv1 <- function(y, shape) {
A8 <- -1 / log1p(-shape)
deriv0 <- A8 * (shape^y) / y
A8 * (shape^(y-1) - deriv0 / (1 - shape))
}
pmf.deriv2 <- function(y, shape) {
A8 <- -1 / log1p(-shape)
A8prime <- -(A8^2) / (1 - shape)
deriv0 <- A8 * (shape^y) / y
deriv1 <- A8 * (shape^(y-1) - deriv0 / (1 - shape))
A8prime * (shape^(y-1) - deriv0 / (1 - shape)) +
A8 * ((y - 1) * shape^(y - 2) - deriv0 / (1 - shape)^2 -
deriv1 / (1 - shape))
}
sumD.mix.1a.p <- sumD.mix.2a.p <- matrix(0, n, NOS)
if (la.mix > 0) { # \calA_p
DA.mix.0mat.a <- # Matches naming convention further below
DA.mix.1mat.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
sumD.mix.1a.p <- sumD.mix.1a.p + pmf.deriv1(aval, shape.p)
sumD.mix.2a.p <- sumD.mix.2a.p + pmf.deriv2(aval, shape.p)
pmf.a <- dlog(aval, shape.a)
DA.mix.0mat.a[, jay] <- pmf.a
DA.mix.1mat.a[, jay] <- pmf.deriv1(aval, shape.a)
}
Denom1.a <- rowSums(DA.mix.1mat.a) # aka sumD.mix.1a.a
} # la.mix > 0
if (li.mix) {
DI.mix.0mat.i <- # wrt inflated distribution
DI.mix.1mat.i <- DI.mix.2mat.i <- matrix(0, n, li.mix)
DP.mix.0mat.i <- # wrt parent distribution
DP.mix.1mat.i <- DP.mix.2mat.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.i <- dlog(ival, shape.i)
DI.mix.0mat.i[, jay] <- pmf.i
DI.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.i)
DI.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.i)
pmf.p <- dlog(ival, shape.p)
DP.mix.0mat.i[, jay] <- pmf.p
DP.mix.1mat.i[, jay] <- pmf.deriv1(ival, shape.p)
DP.mix.2mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
Denom1.i <- rowSums(DI.mix.1mat.i)
Denom2.i <- rowSums(DI.mix.2mat.i)
} # li.mix
if (ld.mix) {
DD.mix.0mat.d <- # wrt deflated distribution
DD.mix.1mat.d <- DD.mix.2mat.d <- matrix(0, n, ld.mix)
DP.mix.0mat.d <- # wrt parent distribution
DP.mix.1mat.d <- DP.mix.2mat.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.d <- dlog(dval, shape.d)
DD.mix.0mat.d[, jay] <- pmf.d
DD.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.d)
DD.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.d)
pmf.p <- dlog(dval, shape.p)
DP.mix.0mat.d[, jay] <- pmf.p
DP.mix.1mat.d[, jay] <- pmf.deriv1(dval, shape.p)
DP.mix.2mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
Denom1.d <- rowSums(DD.mix.1mat.d)
Denom2.d <- rowSums(DD.mix.2mat.d)
} # ld.mix
Bits <- moments.gaitdcombo.log(shape.p,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
a.mix = a.mix, i.mix = i.mix, d.mix = d.mix,
a.mlm = a.mlm, i.mlm = i.mlm, d.mlm = d.mlm,
shape.a = shape.a, shape.i = shape.i,
shape.d = shape.d,
truncate = truncate, max.support = max.support)
sumD.mlm.1a.p <- sumD.mlm.2a.p <- matrix(0, n, NOS)
if (la.mlm)
for (aval in a.mlm) {
sumD.mlm.1a.p <- sumD.mlm.1a.p + pmf.deriv1(aval, shape.p)
sumD.mlm.2a.p <- sumD.mlm.2a.p + pmf.deriv2(aval, shape.p)
}
Denom0.p <- c(Bits[["cdf.max.s"]] - Bits[["SumT0.p"]] -
Bits[["SumA0.mix.p"]] - Bits[["SumA0.mlm.p"]])
Numer <- 1 - pobs.mix - pstr.mix - prob.mlm.a - prob.mlm.i +
pdip.mix + prob.mlm.d
Denom0.a <- c(Bits[["SumA0.mix.a"]]) # Not .p
Denom0.i <- c(Bits[["SumI0.mix.i"]])
Denom0.d <- c(Bits[["SumD0.mix.d"]])
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, NOS)
if (li.mlm > 0) {
Dp.mlm.0Mat.i <- # wrt parent distribution
Dp.mlm.1Mat.i <- Dp.mlm.2Mat.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
pmf.p <- dlog(ival, shape.p)
Dp.mlm.0Mat.i[, jay] <- pmf.p
Dp.mlm.1Mat.i[, jay] <- pmf.deriv1(ival, shape.p)
Dp.mlm.2Mat.i[, jay] <- pmf.deriv2(ival, shape.p)
} # jay
} # li.mlm
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, NOS)
if (ld.mlm > 0) {
Dp.mlm.0Mat.d <- # wrt parent distribution
Dp.mlm.1Mat.d <- Dp.mlm.2Mat.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
pmf.p <- dlog(dval, shape.p)
Dp.mlm.0Mat.d[, jay] <- pmf.p
Dp.mlm.1Mat.d[, jay] <- pmf.deriv1(dval, shape.p)
Dp.mlm.2Mat.d[, jay] <- pmf.deriv2(dval, shape.p)
} # jay
} # ld.mlm
sumD.1t.p <- sumD.2t.p <-
sumD.1t.a <- sumD.2t.a <-
sumD.1t.i <- sumD.2t.i <-
sumD.1t.d <- sumD.2t.d <- matrix(0, n, NOS)
if (ltruncat)
for (tval in truncate) {
sumD.1t.p <- sumD.1t.p + pmf.deriv1(tval, shape.p)
sumD.2t.p <- sumD.2t.p + pmf.deriv2(tval, shape.p)
sumD.1t.a <- sumD.1t.a + pmf.deriv1(tval, shape.a)
sumD.2t.a <- sumD.2t.a + pmf.deriv2(tval, shape.a)
sumD.1t.i <- sumD.1t.i + pmf.deriv1(tval, shape.i)
sumD.2t.i <- sumD.2t.i + pmf.deriv2(tval, shape.i)
sumD.1t.d <- sumD.1t.d + pmf.deriv1(tval, shape.d)
sumD.2t.d <- sumD.2t.d + pmf.deriv2(tval, shape.d)
}
if (is.finite(max.support)) {
tmp1.p <- A8.p * (shape.p^max.support -
(1 - plog(max.support, shape.p))) / (1 - shape.p)
sumD.1t.p <- sumD.1t.p + tmp1.p
sumD.2t.p <- sumD.2t.p + (A8.p / (1 - shape.p)) * (
(shape.p^max.support) / (1 - shape.p) +
max.support * shape.p^(max.support - 1) -
(1 - plog(max.support, shape.p)) / (1 - shape.p) - 2 * tmp1.p)
tmp1.a <- A8.a * (shape.a^max.support -
(1 - plog(max.support, shape.a))) / (1 - shape.a)
sumD.1t.a <- sumD.1t.a + tmp1.a
sumD.2t.a <- sumD.2t.a + (A8.a / (1 - shape.a)) * (
(shape.a^max.support) / (1 - shape.a) +
max.support * shape.a^(max.support - 1) -
(1 - plog(max.support, shape.a)) / (1 - shape.a) - 2 * tmp1.a)
tmp1.i <- A8.i * (shape.i^max.support -
(1 - plog(max.support, shape.i))) / (1 - shape.i)
sumD.1t.i <- sumD.1t.i + tmp1.i
sumD.2t.i <- sumD.2t.i + (A8.i / (1 - shape.i)) * (
(shape.i^max.support) / (1 - shape.i) +
max.support * shape.i^(max.support - 1) -
(1 - plog(max.support, shape.i)) / (1 - shape.i) - 2 * tmp1.i)
} # is.finite(max.support)
Denom1.p <- c(-sumD.1t.p - sumD.mlm.1a.p - sumD.mix.1a.p)
Denom2.p <- c(-sumD.2t.p - sumD.mlm.2a.p - sumD.mix.2a.p)
d0B.PI.mlm <- Dp.mlm.0Mat.i / Denom0.p
d1B.PI.mlm <- Dp.mlm.1Mat.i / Denom0.p - # This is most general
Dp.mlm.0Mat.i * Denom1.p / Denom0.p^2
d2B.PI.mlm <- Dp.mlm.2Mat.i / Denom0.p -
2 * Dp.mlm.1Mat.i * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.i * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.i * (Denom1.p^2) / Denom0.p^3
d0B.PD.mlm <- Dp.mlm.0Mat.d / Denom0.p
d1B.PD.mlm <- Dp.mlm.1Mat.d / Denom0.p - # This is most general
Dp.mlm.0Mat.d * Denom1.p / Denom0.p^2
d2B.PD.mlm <- Dp.mlm.2Mat.d / Denom0.p -
2 * Dp.mlm.1Mat.d * Denom1.p / Denom0.p^2 -
Dp.mlm.0Mat.d * Denom2.p / Denom0.p^2 +
2 * Dp.mlm.0Mat.d * (Denom1.p^2) / Denom0.p^3
DELTA.i.mlm <- if (li.mlm > 0) {
Numer * d0B.PI.mlm + pstr.mlm # n x li.mlm.
} else {
matrix(0, n, 1) # If li.mlm == 0, for rowSums().
}
DELTA.d.mlm <- if (ld.mlm > 0) {
Numer * d0B.PD.mlm - pdip.mlm # n x ld.mlm.
} else {
matrix(0, n, 1) # If ld.mlm == 0, for rowSums().
}
if (li.mix > 0) {
d0A.i <- DI.mix.0mat.i / Denom0.i
d0B.PI.mix <- DP.mix.0mat.i / Denom0.p
DELTA.i.mix <- Numer * d0B.PI.mix + pstr.mix * d0A.i
d1A.i <- (DI.mix.1mat.i - DI.mix.0mat.i *
Denom1.i / Denom0.i) / Denom0.i
d2A.i <- (DI.mix.2mat.i - (2 * DI.mix.1mat.i * Denom1.i +
DI.mix.0mat.i * Denom2.i) / Denom0.i +
2 * DI.mix.0mat.i * (Denom1.i / Denom0.i)^2) / Denom0.i
d1B.PI.mix <- DP.mix.1mat.i / Denom0.p -
DP.mix.0mat.i * Denom1.p / Denom0.p^2
d2B.PI.mix <- DP.mix.2mat.i / Denom0.p -
2 * DP.mix.1mat.i * Denom1.p / Denom0.p^2 -
DP.mix.0mat.i * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.i * (Denom1.p^2) / Denom0.p^3
} # li.mix > 0
if (ld.mix > 0) {
d0A.d <- DD.mix.0mat.d / Denom0.d
d0B.PD.mix <- DP.mix.0mat.d / Denom0.p
DELTA.d.mix <- Numer * d0B.PD.mix - pdip.mix * d0A.d
d1A.d <- (DD.mix.1mat.d - DD.mix.0mat.d *
Denom1.d / Denom0.d) / Denom0.d
d2A.d <- (DD.mix.2mat.d - (2 * DD.mix.1mat.d * Denom1.d +
DD.mix.0mat.d * Denom2.d) / Denom0.d +
2 * DD.mix.0mat.d * (Denom1.d / Denom0.d)^2) / Denom0.d
d1B.PD.mix <- DP.mix.1mat.d / Denom0.p -
DP.mix.0mat.d * Denom1.p / Denom0.p^2
d2B.PD.mix <- DP.mix.2mat.d / Denom0.p -
2 * DP.mix.1mat.d * Denom1.p / Denom0.p^2 -
DP.mix.0mat.d * Denom2.p / Denom0.p^2 +
2 * DP.mix.0mat.d * (Denom1.p^2) / Denom0.p^3
} # ld.mix > 0
if (la.mix) {
d0A.a <- DA.mix.0mat.a / Denom0.a
d1A.a <- DA.mix.1mat.a / Denom0.a -
DA.mix.0mat.a * Denom1.a / Denom0.a^2
} # la.mix
dl.dshape.p <- -A8.p / (1 - shape.p) + y / shape.p
dl.dshape.p[!is.ns] <- 0 # For is.a.mixed & is.a.mlmed
dl.dshape.a <- dl.dshape.i <- dl.dshape.d <- numeric(n)
dl.dpstr.mix <- (-1) / Numer # \notin A, I, T, D
dl.dpstr.mix[is.a.mixed] <- 0
dl.dpstr.mix[is.a.mlmed] <- 0
dl.dpdip.mix <- (+1) / Numer # \notin A, I, T, D
dl.dpdip.mix[is.a.mixed] <- 0
dl.dpdip.mix[is.a.mlmed] <- 0
dl.dpobs.mix <- numeric(n) # 0 for \calA_{np}
dl.dpobs.mix[is.ns] <- (-1) / Numer[is.ns]
dl.dpobs.mlm <-
dl.dpstr.mlm <- matrix(0, n, 1) # May be unneeded
dl.dpdip.mlm <- matrix(0, n, max(1, ld.mlm)) # Initzed if used.
dl.dpdip.mlm[is.ns, ] <- 1 / Numer[is.ns]
if (tmp3.TF[ 8] && la.mlm) { # aka \calA_{np}
dl.dpobs.mlm <- matrix(-1 / Numer, n, la.mlm) # \notin calS
dl.dpobs.mlm[!is.ns, ] <- 0 # For a.mix only really
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
is.alt.j.mlm <- extra$skip.mlm.a[, jay] # Logical vector
tmp7a <- 1 / pobs.mlm[is.alt.j.mlm, jay]
dl.dpobs.mlm[is.alt.j.mlm, jay] <- tmp7a
} # jay
} # la.mlm
dl.dshape.p[is.ns] <- dl.dshape.p[is.ns] -
(Denom1.p / Denom0.p)[is.ns]
if (tmp3.TF[ 9] && li.mlm > 0) { # aka \calI_{np}
dl.dpstr.mlm <- matrix(-1 / Numer, n, li.mlm)
dl.dpstr.mlm[!is.ns, ] <- 0 # For a.mlm and a.mix
for (jay in seq(li.mlm)) {
is.inf.j.mlm <- extra$skip.mlm.i[, jay] # Logical vector
tmp7i <- Numer * d1B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
dl.dshape.p[is.inf.j.mlm] <- tmp7i[is.inf.j.mlm]
tmp9i <- d0B.PI.mlm[, jay] / DELTA.i.mlm[, jay]
n.tmp <- -tmp9i[is.inf.j.mlm]
p.tmp <- +tmp9i[is.inf.j.mlm]
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.inf.j.mlm ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mlm, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mlm ] <- p.tmp
tmp8 <- (1 - d0B.PI.mlm[, jay]) / DELTA.i.mlm[, jay]
dl.dpstr.mlm[is.inf.j.mlm, ] <- n.tmp # tmp9[is.inf.j.mlm]
dl.dpstr.mlm[is.inf.j.mlm, jay] <- tmp8[is.inf.j.mlm]
} # jay
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # aka \calD_{np}
for (jay in seq(ld.mlm)) {
is.def.j.mlm <- extra$skip.mlm.d[, jay] # Logical vector
tmp7d <- Numer * d1B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
dl.dshape.p[is.def.j.mlm] <- tmp7d[is.def.j.mlm] # 20211020
tmp9d <- d0B.PD.mlm[, jay] / DELTA.d.mlm[, jay]
p.tmp <- +tmp9d[is.def.j.mlm]
n.tmp <- -tmp9d[is.def.j.mlm]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mlm, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mlm ] <- n.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.def.j.mlm ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, ] <- p.tmp
dl.dpdip.mlm[is.def.j.mlm, jay] <-
dl.dpdip.mlm[is.def.j.mlm, jay] -
1 / DELTA.d.mlm[is.def.j.mlm, jay]
} # jay
} # ld.mlm > 0
if (tmp3.TF[ 2] && la.mix) { # aka \calA_{p}
dl.dpobs.mix[is.a.mixed] <- 1 / pobs.mix[is.a.mixed]
if (tmp3.TF[ 3] && la.mix > 1)
for (jay in seq(la.mix)) {
is.alt.j.mix <- extra$skip.mix.a[, jay] # Logical vector
tmp2 <- d1A.a[, jay] / d0A.a[, jay]
dl.dshape.a[is.alt.j.mix] <- tmp2[is.alt.j.mix] # ccc.
} # jay
} # la.mix
if (tmp3.TF[ 4] && li.mix > 0) { # aka \calI_{p}
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
is.inf.j.mix <- extra$skip.mix.i[, jay] # Logical vector
tmp7b <- Numer * d1B.PI.mix[, jay] / DELTA.i.mix[, jay]
dl.dshape.p[is.inf.j.mix] <- tmp7b[is.inf.j.mix]
tmp8 <- (d0A.i[, jay] - d0B.PI.mix[, jay]) / DELTA.i.mix[, jay]
dl.dpstr.mix[is.inf.j.mix] <- tmp8[is.inf.j.mix]
if (li.mix > 1) {
tmp2 <- pstr.mix * d1A.i[, jay] / DELTA.i.mix[, jay]
dl.dshape.i[is.inf.j.mix] <- tmp2[is.inf.j.mix]
}
tmp9i <- d0B.PI.mix[, jay] / DELTA.i.mix[, jay]
n.tmp <- -tmp9i[is.inf.j.mix]
p.tmp <- +tmp9i[is.inf.j.mix]
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.inf.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.inf.j.mix, ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.inf.j.mix, ] <- p.tmp
if (tmp3.TF[ 6] && ld.mix) dl.dpdip.mix[is.inf.j.mix ] <- p.tmp
} # jay
} # li.mix > 0
if (tmp3.TF[ 6] && ld.mix > 0) { # aka \calD_{p}
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
is.def.j.mix <- extra$skip.mix.d[, jay] # Logical vector
tmp7b <- Numer * d1B.PD.mix[, jay] / DELTA.d.mix[, jay]
dl.dshape.p[is.def.j.mix] <- tmp7b[is.def.j.mix]
tmp8 <- (d0B.PD.mix[, jay] - d0A.d[, jay]) / DELTA.d.mix[, jay]
dl.dpdip.mix[is.def.j.mix] <- tmp8[is.def.j.mix]
if (ld.mix > 1) {
tmp2 <- (-pdip.mix) * d1A.d[, jay] / DELTA.d.mix[, jay]
dl.dshape.d[is.def.j.mix] <- tmp2[is.def.j.mix]
}
tmp9d <- d0B.PD.mix[, jay] / DELTA.d.mix[, jay]
n.tmp <- -tmp9d[is.def.j.mix]
p.tmp <- +tmp9d[is.def.j.mix]
if (tmp3.TF[ 9] && li.mlm) dl.dpstr.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 4] && li.mix) dl.dpstr.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[ 8] && la.mlm) dl.dpobs.mlm[is.def.j.mix, ] <- n.tmp
if (tmp3.TF[ 2] && la.mix) dl.dpobs.mix[is.def.j.mix ] <- n.tmp
if (tmp3.TF[10] && ld.mlm) dl.dpdip.mlm[is.def.j.mix, ] <- p.tmp
} # jay
} # ld.mix > 0
new.ansd <- matrix(0, n, M) # Same dimension as eta
tmp3.TF <- !is.na(rowSums(extra$indeta))
if (lall.len) { # An MLM fitted
all6.dldp <- cbind(if (tmp3.TF[ 2]) dl.dpobs.mix else NULL,
if (tmp3.TF[ 4]) dl.dpstr.mix else NULL,
if (tmp3.TF[ 6]) dl.dpdip.mix else NULL,
if (tmp3.TF[ 8]) dl.dpobs.mlm else NULL,
if (tmp3.TF[ 9]) dl.dpstr.mlm else NULL,
if (tmp3.TF[10]) dl.dpdip.mlm else NULL)
rSs.tmp <- rowSums(allprobs[, -ncol(allprobs), drop = FALSE] *
all6.dldp)
new.ansd[, -ind.shape.z] <- allprobs[, -ncol(allprobs)] *
(all6.dldp - rSs.tmp)
} # lall.len
dshape.p.deta <- dtheta.deta(shape.p, .lshape.p , .eshape.p )
if (tmp3.TF[ 3])
dshape.a.deta <- dtheta.deta(shape.a, .lshape.a , .eshape.a )
if (tmp3.TF[ 5])
dshape.i.deta <- dtheta.deta(shape.i, .lshape.i , .eshape.i )
if (tmp3.TF[ 7])
dshape.d.deta <- dtheta.deta(shape.d, .lshape.d , .eshape.d )
new.ansd[, 1] <- dl.dshape.p * dshape.p.deta
if (tmp3.TF[ 3])
new.ansd[, extra$indeta[3, 1]] <- dl.dshape.a * dshape.a.deta
if (tmp3.TF[ 5])
new.ansd[, extra$indeta[5, 1]] <- dl.dshape.i * dshape.i.deta
if (tmp3.TF[ 7])
new.ansd[, extra$indeta[7, 1]] <- dl.dshape.d * dshape.d.deta
onecoln.indeta <- extra$indeta[1:7, ] # One coln params only
onecoln.indeta <- na.omit(onecoln.indeta) # Only those present
allcnames <- c(rownames(onecoln.indeta),
as.character(c(a.mlm, i.mlm, d.mlm)))
colnames(new.ansd) <- allcnames
c(w) * new.ansd
}), list(
.lshape.p = lshape.p, .eshape.p = eshape.p,
.lshape.a = lshape.a, .eshape.a = eshape.a,
.lshape.i = lshape.i, .eshape.i = eshape.i,
.lshape.d = lshape.d, .eshape.d = eshape.d,
.lpstr.mix = lpstr.mix, .lpobs.mix = lpobs.mix,
.lpdip.mix = lpdip.mix,
.epstr.mix = epstr.mix, .epobs.mix = epobs.mix,
.epdip.mix = epdip.mix,
.a.mix = a.mix, .i.mix = i.mix, .d.mix = d.mix,
.a.mlm = a.mlm, .i.mlm = i.mlm, .d.mlm = d.mlm,
.tmp3.TF = tmp3.TF, # .tmp3 = tmp3,
.truncate = truncate, .max.support = max.support ))),
weight = eval(substitute(expression({ # gaitdlog
wz <- matrix(0, n, M * (M + 1) / 2) # The complete size
mean.true.p <- A8.p * shape.p / (1 - shape.p)
cond.EY.p <- c(mean.true.p - Bits[["SumT1.p"]] -
Bits[["SumI1.mlm.p"]] - Bits[["SumI1.mix.p"]] -
Bits[["SumD1.mlm.p"]] - Bits[["SumD1.mix.p"]] - # 20211109
Bits[["SumA1.mlm.p"]] - Bits[["SumA1.mix.p"]]) / c(
Denom0.p -
Bits[["SumD0.mix.p"]] - Bits[["SumD0.mlm.p"]] - # 20211109
Bits[["SumI0.mix.p"]] - Bits[["SumI0.mlm.p"]])
probns <- Numer * (1 -
(c(Bits[["SumI0.mix.p"]] + Bits[["SumI0.mlm.p"]]) +
c(Bits[["SumD0.mix.p"]] + Bits[["SumD0.mlm.p"]])) / Denom0.p)
if (min(probns) < 0 || 1 < max(probns))
stop("variable 'probns' for P(nonspecial) is out of range")
zero0n <- numeric(n)
ned2l.dpobs.mix.shape.p <- zero0n # mB overwritten below [4279]
ned2l.dpobs.mix.shape.a <- zero0n # Fini; (2, 3) element
ned2l.dpobs.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpobs.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.p <- zero0n # Optional (1, 4) element
ned2l.dpstr.mix.shape.a <- zero0n # Final; nothing to do
ned2l.dpstr.mix.shape.i <- zero0n # mB overwritten below
ned2l.dpstr.mix.shape.d <- zero0n # mB overwritten below
ned2l.dpdip.mix.shape.p <- zero0n # Optional (1, 6) element
posn.pobs.mix <- as.vector(extra$indeta[ 2, 'launch'])
posn.shape.a <- as.vector(extra$indeta[ 3, 'launch'])
posn.pstr.mix <- as.vector(extra$indeta[ 4, 'launch'])
posn.shape.i <- as.vector(extra$indeta[ 5, 'launch'])
posn.pdip.mix <- as.vector(extra$indeta[ 6, 'launch'])
posn.shape.d <- as.vector(extra$indeta[ 7, 'launch'])
posn.pobs.mlm <- as.vector(extra$indeta[ 8, 'launch'])
posn.pstr.mlm <- as.vector(extra$indeta[ 9, 'launch'])
posn.pdip.mlm <- as.vector(extra$indeta[10, 'launch'])
ned2l.dpdip.mix2 <- # Elt (6, 6)
ned2l.dpstr.mix2 <- # Elt (4, 4). Unchanged by deflation.
ned2l.dpobs.mlm.pstr.mix <- # Elts (4, >=8). (((09)))
ned2l.dpobs.mix.pstr.mix <- +probns / Numer^2 # ccc Elt (2, 4)
if (all(c(la.mix, li.mlm) > 0)) # (((08)))
ned2l.dpobs.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(li.mix, li.mlm) > 0)) # (((10)))
ned2l.dpstr.mix.pstr.mlm <- matrix( probns / Numer^2, n, li.mlm)
if (all(c(ld.mix, ld.mlm) > 0)) # (((21)))
ned2l.dpdip.mix.pdip.mlm <- matrix( probns / Numer^2, n, ld.mlm)
ned2l.dpobs.mlm.pdip.mix <- # Elts (6, >=8). (((19)))
ned2l.dpstr.mix.pdip.mix <- # Elt (4, 6)
ned2l.dpobs.mix.pdip.mix <- -probns / Numer^2 # ccc Elt (2, 6)
if (all(c(la.mix, ld.mlm) > 0)) # (((17)))
ned2l.dpobs.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(li.mix, ld.mlm) > 0)) # (((18)))
ned2l.dpstr.mix.pdip.mlm <- matrix(-probns / Numer^2, n, ld.mlm)
if (all(c(ld.mix, li.mlm) > 0)) # (((20)))
ned2l.dpdip.mix.pstr.mlm <- matrix(-probns / Numer^2, n, li.mlm)
ned2l.dshape.p2 <- probns * (cond.EY.p / shape.p^2 + # ccc
A8.p * (1 - A8.p) / (1 - shape.p)^2 +
Denom2.p / Denom0.p - (Denom1.p / Denom0.p)^2) +
(if (tmp3.TF[ 4] && li.mix) Numer *
rowSums(Numer * (d1B.PI.mix^2) / DELTA.i.mix - d2B.PI.mix)
else 0) +
(if (tmp3.TF[ 9] && li.mlm) Numer *
rowSums(Numer * (d1B.PI.mlm^2) / DELTA.i.mlm - d2B.PI.mlm)
else 0) +
(if (tmp3.TF[ 6] && ld.mix) Numer *
rowSums(Numer * (d1B.PD.mix^2) / DELTA.d.mix - d2B.PD.mix)
else 0) +
(if (tmp3.TF[10] && ld.mlm) Numer * # nnn.
rowSums(Numer * (d1B.PD.mlm^2) / DELTA.d.mlm - d2B.PD.mlm)
else 0)
wz[, iam(1, 1, M)] <- ned2l.dshape.p2 * dshape.p.deta^2
ned2l.dpobs.mix2 <- 1 / pobs.mix + probns / Numer^2
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpobs.mix2 <- # More just below, ccc
ned2l.dpobs.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpobs.mix2 <- # ccc.
ned2l.dpobs.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (tmp3.TF[ 6] && ld.mix > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpobs.mix2 <- # nnn
ned2l.dpobs.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (tmp3.TF[ 2] && la.mix > 0)
wz[, iam(2, 2, M)] <- ned2l.dpobs.mix2 # Link done later
if (tmp3.TF[ 3] && la.mix > 1) {
ned2l.dshape.a2 <- pobs.mix * (
rowSums((DA.mix.1mat.a^2) / DA.mix.0mat.a) / Denom0.a -
(Denom1.a / Denom0.a)^2) # ccc.
wz[, iam(3, 3, M)] <- ned2l.dshape.a2 * dshape.a.deta^2
}
if (tmp3.TF[ 4] && li.mix > 0) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 +
rowSums((d0A.i - d0B.PI.mix)^2 / DELTA.i.mix)
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mix * (1 + Numer * (d0A.i - d0B.PI.mix) / DELTA.i.mix))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (all(tmp3.TF[c(2, 4)]))
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix +
rowSums(-d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PI.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
} # (tmp3.TF[ 4] && li.mix > 0)
if (all(tmp3.TF[c(2, 4, 9)])) { # was la.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pstr.mix <- # ccc
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 4, 6)])) { # == ld.mix > 0 & DELTA.d.mix
ned2l.dpobs.mix.pstr.mix <- # nnn
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(2, 4, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pstr.mix <- # nnn.
ned2l.dpobs.mix.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pstr.mix))
wz[, iam(posn.pobs.mix, posn.pstr.mix, M)] <-
ned2l.dpobs.mix.pstr.mix # Link done later
if (all(tmp3.TF[c(2, 6)]))
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix +
rowSums( d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
if (all(tmp3.TF[c(2, 6, 9)])) { # == li.mlm > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(2, 6, 4)])) { # == li.mix > 0 & DELTA.i.mix
ned2l.dpobs.mix.pdip.mix <- # nnn
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
}
if (all(tmp3.TF[c(2, 6, 10)])) { # == ld.mlm > 0 & DELTA.d.mlm
ned2l.dpobs.mix.pdip.mix <- # nnn.
ned2l.dpobs.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (!is.na(posn.pobs.mix) && !is.na(posn.pdip.mix))
wz[, iam(posn.pobs.mix, posn.pdip.mix, M)] <-
ned2l.dpobs.mix.pdip.mix # Link done later
if (tmp3.TF[ 5] && li.mix > 1) { # \calI_{p}, includes \theta_i.
ned2l.dshape.p.shape.i <- pstr.mix * Numer *
rowSums(d1A.i * d1B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(1, posn.shape.i, M)] <- ned2l.dshape.p.shape.i *
dshape.p.deta * dshape.i.deta # All links done here
ned2l.dshape.i2 <- pstr.mix *
rowSums(pstr.mix * (d1A.i^2) / DELTA.i.mix - d2A.i) # ccc.
wz[, iam(posn.shape.i, posn.shape.i, M)] <-
ned2l.dshape.i2 * dshape.i.deta^2
if (tmp3.TF[ 2]) { # tmp3.TF[ 4] is TRUE, given tmp3.TF[ 5]
ned2l.dpobs.mix.shape.i <-
rowSums(-pstr.mix * d1A.i * d0B.PI.mix / DELTA.i.mix) # ccc.
wz[, iam(posn.pobs.mix, posn.shape.i, M)] <-
ned2l.dpobs.mix.shape.i # * dshape.i.deta done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.i <- rowSums( # ccc.
d1A.i * (pstr.mix * (d0A.i - d0B.PI.mix) / DELTA.i.mix - 1))
wz[, iam(posn.pstr.mix, posn.shape.i, M)] <-
ned2l.dpstr.mix.shape.i # * dshape.i.deta done later
}
if (all(tmp3.TF[c(5, 6)])) {
ned2l.dpdip.mix.shape.i <- rowSums(
(-pstr.mix) * d0B.PI.mix * d1A.i / DELTA.i.mix)
wz[, iam(posn.pdip.mix, posn.shape.i, M)] <-
ned2l.dpdip.mix.shape.i # link done later
}
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.i <- rowSums(
-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix) # ccc.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.i, M)] <-
ned2l.dpobs.mlm.shape.i # * dshape.i.deta done later
}
} # (tmp3.TF[ 5] && li.mix > 1)
if (tmp3.TF[ 6] && ld.mix > 0) { # \calD_{p}, maybe w. \theta_d
if (tmp3.TF[ 2] && la.mix > 0)
ned2l.dpobs.mix.shape.p <-
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpstr.mix.shape.p <-
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mix * (1 + Numer * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix + rowSums(
d0B.PD.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix)
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 +
rowSums((d0A.d - d0B.PD.mix)^2 / DELTA.d.mix)
} # (tmp3.TF[ 6] && ld.mix > 0)
if (tmp3.TF[ 7] && ld.mix > 1) { # \calD_{p}, includes \theta_d
ned2l.dshape.p.shape.d <- (-pdip.mix) * Numer *
rowSums(d1A.d * d1B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(1, posn.shape.d, M)] <- ned2l.dshape.p.shape.d *
dshape.p.deta * dshape.d.deta # All links done here
if (tmp3.TF[ 2]) { # tmp3.TF[ 6] is TRUE, given tmp3.TF[ 7]
ned2l.dpobs.mix.shape.d <-
rowSums(pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix) # nnn.
wz[, iam(posn.pobs.mix, posn.shape.d, M)] <-
ned2l.dpobs.mix.shape.d # link done later
}
if (tmp3.TF[ 4]) {
ned2l.dpstr.mix.shape.d <- rowSums(
pdip.mix * d1A.d * d0B.PD.mix / DELTA.d.mix)
wz[, iam(posn.pstr.mix, posn.shape.d, M)] <-
ned2l.dpstr.mix.shape.d # * dshape.i.deta done later
}
ned2l.dpdip.mix.shape.d <- rowSums(
d1A.d * (1 + pdip.mix * (d0A.d - d0B.PD.mix) / DELTA.d.mix))
wz[, iam(posn.pdip.mix, posn.shape.d, M)] <-
ned2l.dpdip.mix.shape.d # * dshape.d.deta done later
ned2l.dshape.d2 <- pdip.mix *
rowSums(pdip.mix * (d1A.d^2) / DELTA.d.mix + d2A.d) # nnn.
wz[, iam(posn.shape.d, posn.shape.d, M)] <-
ned2l.dshape.d2 * dshape.d.deta^2
if (tmp3.TF[ 8]) {
ned2l.dpobs.mlm.shape.d <- rowSums(
pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix) # nnn.
for (uuu in seq(la.mlm))
wz[, iam(posn.pobs.mlm - 1 + uuu, posn.shape.d, M)] <-
ned2l.dpobs.mlm.shape.d # * dshape.d.deta done later
}
} # (tmp3.TF[ 7] && ld.mix > 1)
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # ccc
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ned2l.dpstr.mix.shape.p <- # ccc.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
}
} # tmp3.TF[ 9] && li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s.
if (la.mix && tmp3.TF[ 2])
ned2l.dpobs.mix.shape.p <- # nnn.
ned2l.dpobs.mix.shape.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpstr.mix.shape.p <- # nnn.
ned2l.dpstr.mix.shape.p + rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (tmp3.TF[ 6])
ned2l.dpdip.mix.shape.p <-
ned2l.dpdip.mix.shape.p - rowSums(
d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
if (!is.na(posn.pstr.mix)) {
ned2l.dpstr.mix2 <-
ned2l.dpstr.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
if (all(tmp3.TF[c(4, 6)]))
ned2l.dpstr.mix.pdip.mix <-
ned2l.dpstr.mix.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (!is.na(posn.pdip.mix)) {
ned2l.dpdip.mix2 <-
ned2l.dpdip.mix2 + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
}
} # tmp3.TF[10] && ld.mlm > 0
if (!is.na(posn.pobs.mix)) # Optional (1, 2) element:
wz[, iam(1, posn.pobs.mix, M)] <-
ned2l.dpobs.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix)) # Optional (1, 4) element
wz[, iam(1, posn.pstr.mix, M)] <-
ned2l.dpstr.mix.shape.p # One link done later
if (!is.na(posn.pdip.mix)) # Optional (1, 6) element
wz[, iam(1, posn.pdip.mix, M)] <-
ned2l.dpdip.mix.shape.p # One link done later
if (!is.na(posn.pstr.mix) &&
!is.na(posn.pdip.mix)) # Optional (4, 6) element
wz[, iam(posn.pstr.mix, posn.pdip.mix, M)] <-
ned2l.dpstr.mix.pdip.mix # Links done later zz1
if (!is.na(posn.pstr.mix)) # Optional (4, 4) element
wz[, iam(posn.pstr.mix, # Link done later
posn.pstr.mix, M)] <- ned2l.dpstr.mix2
if (!is.na(posn.pdip.mix)) # Optional (6, 6) element
wz[, iam(posn.pdip.mix, # Link done later
posn.pdip.mix, M)] <- ned2l.dpdip.mix2
if (tmp3.TF[ 8] && la.mlm) { # \calA_{np}, includes \omega_s
ofset <- posn.pobs.mlm - 1 # 7 for GAITD combo
for (uuu in seq(la.mlm)) { # Diagonal elts only
wz[, iam(ofset + uuu,
ofset + uuu, M)] <- 1 / pobs.mlm[, uuu]
} # uuu
tmp8a <- probns / Numer^2
if (tmp3.TF[ 4] && li.mix)
tmp8a <- tmp8a + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
tmp8a <- tmp8a + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
tmp8a <- tmp8a + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
tmp8a <- tmp8a + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # All elts
for (vvv in uuu:la.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
wz[, iam(ofset + uuu, ofset + vvv, M)] + tmp8a # All elts
} # la.mlm
if (tmp3.TF[ 8] && la.mlm) {
init0.i.val <- init0.d.val <- 0
if (tmp3.TF[ 9] && li.mlm) init0.i.val <-
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
if (tmp3.TF[10] && ld.mlm) init0.d.val <-
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ned2l.dpobs.mlm.shape.p <- init0.i.val + init0.d.val # Vector
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums(
d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.shape.p <-
ned2l.dpobs.mlm.shape.p + rowSums( # nnn
d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
ofset <- posn.pobs.mlm - 1 # 5 for combo
for (vvv in seq(la.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpobs.mlm.shape.p
} # la.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) { # \calI_{np}, includes \phi_s
init0.val <- probns / Numer^2
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (ld.mix) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (ld.mlm) # nnn
init0.val <- init0.val + rowSums((d0B.PD.mlm^2) / DELTA.d.mlm)
ned2l.dpstr.mlm2 <-
matrix(init0.val, n, li.mlm * (li.mlm + 1) / 2)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, uuu, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss])^2 / DELTA.i.mlm[, sss]
if (li.mlm > 1) {
for (uuu in seq(li.mlm - 1))
for (vvv in (uuu + 1):li.mlm)
for (sss in seq(li.mlm))
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
} # if (li.mlm > 1)
ofset <- posn.pstr.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in uuu:li.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpstr.mlm2[, iam(uuu, vvv, li.mlm)]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) { # \calD_{np}, includes \psi_s
init0.val <- probns / Numer^2
if (ld.mix)
init0.val <- init0.val + rowSums((d0B.PD.mix^2) / DELTA.d.mix)
if (li.mix)
init0.val <- init0.val + rowSums((d0B.PI.mix^2) / DELTA.i.mix)
if (li.mlm)
init0.val <- init0.val + rowSums((d0B.PI.mlm^2) / DELTA.i.mlm)
ned2l.dpdip.mlm2 <-
matrix(init0.val, n, ld.mlm * (ld.mlm + 1) / 2)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, uuu, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu))^2 / DELTA.d.mlm[, sss]
if (ld.mlm > 1) {
for (uuu in seq(ld.mlm - 1))
for (vvv in (uuu + 1):ld.mlm)
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)] +
(d0B.PD.mlm[, sss] - (sss == uuu)) *
(d0B.PD.mlm[, sss] - (sss == vvv)) / DELTA.d.mlm[, sss]
} # if (ld.mlm > 1)
ofset <- posn.pdip.mlm - 1
for (uuu in seq(ld.mlm))
for (vvv in uuu:ld.mlm)
wz[, iam(ofset + uuu, ofset + vvv, M)] <-
ned2l.dpdip.mlm2[, iam(uuu, vvv, ld.mlm)]
} # ld.mlm > 0
if (tmp3.TF[ 9] && li.mlm > 0) {
ned2l.dpstr.mlm.theta.p <- matrix(0, n, li.mlm)
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.theta.p[, vvv] <-
ned2l.dpstr.mlm.theta.p[, vvv] +
d1B.PI.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PI.mlm[, sss]) / (
DELTA.i.mlm[, sss]))
if (li.mix && tmp3.TF[ 4])
ned2l.dpstr.mlm.theta.p <-
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpstr.mlm.theta.p <- # nnn
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (ld.mlm && tmp3.TF[10])
ned2l.dpstr.mlm.theta.p <- # nnn.
ned2l.dpstr.mlm.theta.p +
rowSums(d1B.PD.mlm * (1 - Numer * d0B.PD.mlm / DELTA.d.mlm))
ofset <- posn.pstr.mlm - 1
for (vvv in seq(li.mlm)) # ccc.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpstr.mlm.theta.p[, vvv]
} # li.mlm > 0
if (tmp3.TF[10] && ld.mlm > 0) {
ned2l.dpdip.mlm.theta.p <- matrix(0, n, ld.mlm)
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpdip.mlm.theta.p[, vvv] <-
ned2l.dpdip.mlm.theta.p[, vvv] - # Minus
d1B.PD.mlm[, sss] * (1 + Numer *
(max(0, sss == vvv) - d0B.PD.mlm[, sss]) / (
DELTA.d.mlm[, sss]))
if (ld.mix && tmp3.TF[ 6])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PD.mix * (1 - Numer * d0B.PD.mix / DELTA.d.mix))
if (li.mix && tmp3.TF[ 4])
ned2l.dpdip.mlm.theta.p <-
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mix * (1 - Numer * d0B.PI.mix / DELTA.i.mix))
if (li.mlm && tmp3.TF[ 9])
ned2l.dpdip.mlm.theta.p <- # nnn.
ned2l.dpdip.mlm.theta.p - # Minus
rowSums(d1B.PI.mlm * (1 - Numer * d0B.PI.mlm / DELTA.i.mlm))
ofset <- posn.pdip.mlm - 1
for (vvv in seq(ld.mlm)) # nnn.
wz[, iam(1, ofset + vvv, M)] <- ned2l.dpdip.mlm.theta.p[, vvv]
} # ld.mlm > 0
if (li.mlm && li.mix > 1) {
ned2l.dpstr.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.i, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.i # ccc.
} # li.mlm && li.mix > 1
if (ld.mlm && ld.mix > 1) {
ned2l.dpdip.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.d, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.d # nnn.
} # ld.mlm && ld.mix > 1
if (ld.mlm && li.mix > 1) {
ned2l.dpdip.mlm.theta.i <- # Not a matrix, just a vector
rowSums(-pstr.mix * d0B.PI.mix * d1A.i / DELTA.i.mix)
for (vvv in seq(ld.mlm))
wz[, iam(posn.shape.i, posn.pdip.mlm - 1 + vvv, M)] <-
ned2l.dpdip.mlm.theta.i # nnn.
} # ld.mlm && li.mix > 1
if (li.mlm && ld.mix > 1) {
ned2l.dpstr.mlm.theta.d <- # Not a matrix, just a vector
rowSums(pdip.mix * d0B.PD.mix * d1A.d / DELTA.d.mix)
for (vvv in seq(li.mlm))
wz[, iam(posn.shape.d, posn.pstr.mlm - 1 + vvv, M)] <-
ned2l.dpstr.mlm.theta.d # nnn.
} # li.mlm && ld.mix > 1
if (all(c(la.mlm, li.mlm) > 0)) {
ned2l.dpobs.mlm.pstr.mlm <-
array(probns / Numer^2, c(n, la.mlm, li.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv] - d0B.PI.mlm[, sss] *
((sss == vvv) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pstr.mlm <-
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (tmp3.TF[10] && ld.mlm)
ned2l.dpobs.mlm.pstr.mlm <- # nnn
ned2l.dpobs.mlm.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pstr <- posn.pstr.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(li.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pstr + vvv, M)] <-
ned2l.dpobs.mlm.pstr.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(li.mlm, ld.mlm) > 0)) {
ned2l.dpstr.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, li.mlm, ld.mlm))
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(li.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PI.mlm[, sss] *
((sss == uuu) - d0B.PI.mlm[, sss]) / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpstr.mlm.pdip.mlm <-
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpstr.mlm.pdip.mlm <- # nnn.
ned2l.dpstr.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pstr <- posn.pstr.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(li.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pstr + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpstr.mlm.pdip.mlm[, uuu, vvv]
} # all(c(li.mlm, ld.mlm) > 0)
if (all(c(la.mlm, ld.mlm) > 0)) {
ned2l.dpobs.mlm.pdip.mlm <-
array(-probns / Numer^2, c(n, la.mlm, ld.mlm))
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv] + d0B.PD.mlm[, sss] *
((sss == vvv) - d0B.PD.mlm[, sss]) / DELTA.d.mlm[, sss]
if (tmp3.TF[ 4] && li.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (tmp3.TF[ 9] && li.mlm)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (tmp3.TF[ 6] && ld.mix)
ned2l.dpobs.mlm.pdip.mlm <-
ned2l.dpobs.mlm.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
ofset.pobs <- posn.pobs.mlm - 1
ofset.pdip <- posn.pdip.mlm - 1
for (uuu in seq(la.mlm))
for (vvv in seq(ld.mlm))
wz[, iam(ofset.pobs + uuu, ofset.pdip + vvv, M)] <-
ned2l.dpobs.mlm.pdip.mlm[, uuu, vvv]
} # all(c(la.mlm, li.mlm) > 0)
if (all(c(la.mix, la.mlm) > 0)) {
ned2l.dpobs.mix.pobs.mlm <- probns / Numer^2 # Initialize
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 7]
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10] nnn
ned2l.dpobs.mix.pobs.mlm <-
ned2l.dpobs.mix.pobs.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pobs.mix, posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pobs.mlm # Link done later
}
if (all(c(la.mix, li.mlm) > 0)) { # all(tmp3.TF[c(2, 9)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pstr.mlm <-
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pstr.mlm <- # nnn
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mix.pstr.mlm <- # nnn; + is correct, not -
ned2l.dpobs.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm))
for (sss in seq(li.mlm))
ned2l.dpobs.mix.pstr.mlm[, uuu] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
for (uuu in seq(li.mlm)) # ccc.
wz[, iam(posn.pobs.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pstr.mlm[, uuu] # Link done later
} # all(c(la.mix, li.mlm) > 0)
if (all(c(la.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(2, 10)])
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mix.pdip.mlm <-
ned2l.dpobs.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
for (uuu in seq(ld.mlm))
for (sss in seq(ld.mlm))
ned2l.dpobs.mix.pdip.mlm[, uuu] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
for (uuu in seq(ld.mlm)) # nnn.
wz[, iam(posn.pobs.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mix.pdip.mlm[, uuu] # Link done later
} # all(c(la.mix, ld.mlm) > 0)
if (all(c(li.mix, la.mlm) > 0)) { # all(tmp3.TF[c(4, 8)])
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pstr.mix <-
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pstr.mix <- # nnn
ned2l.dpobs.mlm.pstr.mix + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
ned2l.dpobs.mlm.pstr.mix <- # tmp3.TF[ 4] && li.mix
ned2l.dpobs.mlm.pstr.mix -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
for (uuu in seq(la.mlm)) # ccc.
wz[, iam(posn.pstr.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pstr.mix # Link done later
} # all(c(li.mix, la.mlm) > 0
if (all(c(ld.mix, la.mlm) > 0)) { # all(tmp3.TF[c(6, 8)])
if (ld.mlm) # tmp3.TF[10]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpobs.mlm.pdip.mix <-
ned2l.dpobs.mlm.pdip.mix - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
ned2l.dpobs.mlm.pdip.mix <- # all(tmp3.TF[c(6, 8)])
ned2l.dpobs.mlm.pdip.mix +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
for (uuu in seq(la.mlm)) # nnn.
wz[, iam(posn.pdip.mix,
posn.pobs.mlm - 1 + uuu, M)] <-
ned2l.dpobs.mlm.pdip.mix # Link done later
} # all(c(ld.mix, la.mlm) > 0
if (all(c(li.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpstr.mix.pstr.mlm[, uuu] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] -
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
ned2l.dpstr.mix.pstr.mlm <-
ned2l.dpstr.mix.pstr.mlm -
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpstr.mix.pstr.mlm <- # nnn
ned2l.dpstr.mix.pstr.mlm + rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pstr.mlm[, uuu] # Link done later
} # all(c(li.mix, li.mlm) > 0
if (all(c(ld.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(6, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[ 9]
for (sss in seq(ld.mlm))
ned2l.dpdip.mix.pdip.mlm[, uuu] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] -
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm -
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpdip.mix.pdip.mlm <-
ned2l.dpdip.mix.pdip.mlm + rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pdip.mlm[, uuu] # Link done later
} # all(c(ld.mix, ld.mlm) > 0
if (all(c(ld.mix, li.mlm) > 0)) { # all(tmp3.TF[c(4, 9)])
for (uuu in seq(li.mlm)) # tmp3.TF[ 9]
for (sss in seq(li.mlm))
ned2l.dpdip.mix.pstr.mlm[, uuu] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] +
((sss == uuu) - d0B.PI.mlm[, sss]) *
d0B.PI.mlm[, sss] / DELTA.i.mlm[, sss]
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm +
rowSums((d0A.d - d0B.PD.mix) * d0B.PD.mix / DELTA.d.mix)
if (li.mix) # tmp3.TF[ 4]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PI.mix^2 / DELTA.i.mix)
if (ld.mlm) # tmp3.TF[10]
ned2l.dpdip.mix.pstr.mlm <-
ned2l.dpdip.mix.pstr.mlm - rowSums(d0B.PD.mlm^2 / DELTA.d.mlm)
for (uuu in seq(li.mlm)) # Copy it. ccc.
wz[, iam(posn.pdip.mix,
posn.pstr.mlm - 1 + uuu, M)] <-
ned2l.dpdip.mix.pstr.mlm[, uuu] # Link done later
} # all(c(ld.mix, li.mlm) > 0
if (all(c(li.mix, ld.mlm) > 0)) { # all(tmp3.TF[c(4, 10)])
for (uuu in seq(ld.mlm)) # tmp3.TF[10]
for (sss in seq(ld.mlm))
ned2l.dpstr.mix.pdip.mlm[, uuu] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] +
((sss == uuu) - d0B.PD.mlm[, sss]) *
d0B.PD.mlm[, sss] / DELTA.d.mlm[, sss]
if (li.mix) # tmp3.TF[ 4]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm +
rowSums((d0A.i - d0B.PI.mix) * d0B.PI.mix / DELTA.i.mix)
if (ld.mix) # tmp3.TF[ 6]
ned2l.dpstr.mix.pdip.mlm <-
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PD.mix^2 / DELTA.d.mix)
if (li.mlm) # tmp3.TF[ 9]
ned2l.dpstr.mix.pdip.mlm <- # nnn.
ned2l.dpstr.mix.pdip.mlm - rowSums(d0B.PI.mlm^2 / DELTA.i.mlm)
for (uuu in seq(ld.mlm)) # Copy it. ccc.
wz[, iam(posn.pstr.mix,
posn.pdip.mlm - 1 + uuu, M)] <-
ned2l.dpstr.mix.pdip.mlm[, uuu] # Link done later
} # all(c(li.mix, ld.mlm) > 0)
if (lall.len) {
wz.6 <- matrix(0, n, M * (M + 1) / 2) # Or == 0 * wz
ind.rc <- setdiff(1:M, ind.shape.z) # Contiguous rows and
lind.rc <- length(ind.rc) # cols of the DAMLM
# Copy in the thetas values: the looping is overkill.
for (uuu in ind.shape.z)
for (sss in seq(M))
wz.6[, iam(uuu, sss, M)] <- wz[, iam(uuu, sss, M)]
speed.up <- intercept.only && (
length(offset) == 1 || all(offset[1] == offset))
IND.mlm <- iam(NA, NA, lind.rc, both = TRUE, diag = TRUE)
n.use <- if (speed.up) 2 else n # For sandwich.mlm
if (!length(extra$ind.wz.match)) {
Imat <- matrix(NA, lind.rc, lind.rc)
for (jay in seq(lind.rc)) {
iptr <- jay
for (kay in (ind.rc[jay]):M) {
if (!any(kay %in% ind.shape.z)) {
Imat[jay, iptr] <-
which(extra$index.M$row == ind.rc[jay] &
extra$index.M$col == kay)
iptr <- iptr + 1
} # if
} # kay
} # jay
ind.wz.match <- Imat[cbind(IND.mlm$row.ind,
IND.mlm$col.ind)]
extra$ind.wz.match <- ind.wz.match # Assign it once
} # !length(extra$ind.wz.match)
filling <- if (speed.up)
wz[1:n.use, extra$ind.wz.match, drop = FALSE] else
wz[, extra$ind.wz.match, drop = FALSE]
M.mlm <- lind.rc
if (is.null(extra$iamlist)) {
extra$iamlist <- iamlist <-
iam(NA, NA, M = M.mlm, both = TRUE)
if (M.mlm > 1) { # Offdiagonal elts
extra$iamlist.nod <- iamlist.nod <-
iam(NA, NA, M.mlm, both = TRUE, diag = FALSE)
}
} # is.null(extra$iamlist)
iamlist <- extra$iamlist
iamlist.nod <- extra$iamlist.nod
MM12.mlm <- M.mlm * (M.mlm + 1) / 2
Qf3 <- rowSums(filling[, 1:M.mlm, drop = FALSE] * # Diag elts
(allprobs[1:n.use, 1:M.mlm, drop = FALSE])^2)
if (M.mlm > 1) # Offdiagonal elts
Qf3 <- Qf3 + 2 * rowSums(allprobs[1:n.use, iamlist.nod$row] *
filling[, -(1:M.mlm), drop = FALSE] * # n-vector
allprobs[1:n.use, iamlist.nod$col])
Qf3 <- matrix(Qf3, n.use, MM12.mlm)
Qf2rowsums <- matrix(0, n.use, M.mlm) # rowsums stored columnwise
for (want in seq(M.mlm)) { # Want the equivalent of rowSums(Qf2a)
iamvec <- iam(want, 1:M.mlm, M = M.mlm) # Diagonals included
Qf2rowsums[, want] <- rowSums(filling[, iamvec, drop = FALSE] *
allprobs[1:n.use, 1:M.mlm])
} # want
Qf2a <- Qf2rowsums[, iamlist$row]
Qf2b <- Qf2rowsums[, iamlist$col]
Qform <- filling - Qf2a - Qf2b + Qf3 # n x MM12.mlm
Qform <- Qform *
allprobs[1:n.use, iamlist$row, drop = FALSE] *
allprobs[1:n.use, iamlist$col, drop = FALSE]
wz.6[, extra$ind.wz.match] <- if (speed.up)
matrix(Qform[1, ], n, ncol(Qform), byrow = TRUE) else c(Qform)
dstar.deta <- cbind(dshape.p.deta,
if (tmp3.TF[ 3]) dshape.a.deta else NULL,
if (tmp3.TF[ 5]) dshape.i.deta else NULL,
if (tmp3.TF[ 7]) dshape.d.deta else NULL)
iptr <- 0
if (length(ind.shape.z))
for (uuu in ind.shape.z) { # Could delete 3 for shape.a (orthog)
iptr <- iptr + 1
for (ttt in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <- 0 # Initialize
for (sss in seq(lind.rc)) {
wz.6[, iam(uuu, ind.rc[ttt], M)] <-
wz.6[, iam(uuu, ind.rc[ttt], M)] +
allprobs[, sss] * (max(0, sss == ttt) - allprobs[, ttt]) *
wz[, iam(uuu, ind.rc[sss], M)] * dstar.deta[, iptr]
} # sss
} # ttt
} # uuu
wz <- wz.6 # Completed
} # lall.len
if (lall.len) { # A MLM was fitted
mytiny <- (allprobs < sqrt(.Machine$double.eps)) |
(allprobs > 1.0 - sqrt(.Machine$double.eps))
atiny <- rowSums(mytiny) > 0
if (any(atiny)) {
ind.diags <- setdiff(1:M, ind.shape.z) # Exclude thetas
wz[atiny, ind.diags] <- .Machine$double.eps +
wz[atiny, ind.diags] * (1 + .Machine$double.eps^0.5)
}
} # lall.len
c(w) * wz
}), list( .truncate = truncate ))))
} # gaitdlog
moments.gaitdcombo.binom <-
function(size.p, prob.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
size.a = size.p,
size.i = size.p,
size.d = size.p,
prob.a = prob.p,
prob.i = prob.p,
prob.d = prob.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
if (is.infinite(max.support)) {
rmlife1 <- rmlife2 <- numeric(length(size.p)) # 0
} else {
stop("currently RML unknown for finite 'max.support'")
x.use <- max.support + 1
rmlife1 <- NA
rmlife2 <- NA
} # is.infinite(max.support)
mylist1 <-
moments.gaitdcombo.2par(
theta1.p = size.p, theta2.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta1.a = size.a, theta2.a = prob.a,
theta1.i = size.i, theta2.i = prob.i,
theta1.d = size.d, theta2.d = prob.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dgaitdbinom") # do.call() called.
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this * (munb.p - SumA1.mix.p -
SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.binom
dgaitdbinom <-
function(x, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, # vector
pobs.mlm = 0, # matrix
pstr.mix = 0, # vector
pstr.mlm = 0, # matrix
pdip.mix = 0, # vector
pdip.mlm = 0, # matrix
byrow.aid = FALSE, # Applies to 'pobs.mlm' & 'pstr.mlm'
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
log = FALSE,
...) { # ... is for max.support (ignored)
max.support <- Inf
log.arg <- log; rm(log)
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0 # Lower support
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(dbinom(x, size.p, prob.p, log = log.arg))
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(x),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a),
length(size.i), length(size.d),
length(prob.p), length(prob.a),
length(prob.i), length(prob.d))
if (length(x) < LLL) x <- rep_len(x, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
sumt <- 0 # Initialization to 0 important
if (ltrunc)
for (tval in truncate)
sumt <- sumt + dbinom(tval, size.p, prob.p)
vecTF.t <- is.finite(x) & ((x %in% truncate) | (max.support < x))
cdf.max.s <- pbinom(max.support, size.p, prob.p) # Usually 1
denom.t <- cdf.max.s - sumt # No sumt on RHS
pmf0 <- ifelse(vecTF.t, 0, dbinom(x, size.p, prob.p) / denom.t)
sum.a <- suma <- 0 # numeric(LLL)
vecTF.a <- rep_len(FALSE, LLL)
if (la.mlm) {
pobs.mlm <- matrix(pobs.mlm, LLL, la.mlm, byrow = byrow.aid)
sum.a <- .rowSums(pobs.mlm, LLL, la.mlm)
if (any(1 < sum.a, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'") # zz
for (aval in a.mlm)
suma <- suma + dbinom(aval, size.p, prob.p) # Part i
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
if (any(vecTF <- is.finite(x) & aval == x)) {
pmf0[vecTF] <- pobs.mlm[vecTF, jay]
}
vecTF.a <- vecTF.a | vecTF # Cumulative
} # jay
} # la.mlm
pmf2.a <- pmf2.i <- pmf2.d <- 0
if (la.mix) {
allx.a <- lowsup:max(a.mix)
pmf2.a <- dgaitdbinom(x, # Outer distribution---mlm type
size.a, prob.a,
truncate = setdiff(allx.a, a.mix))
for (aval in a.mix) { # Part ii added; cumulative
suma <- suma + dbinom(aval, size.p, prob.p)
vecTF <- is.finite(x) & aval == x
pmf0[vecTF] <- 0 # added; the true values are assigned below
vecTF.a <- vecTF.a | vecTF # Cumulative; added
}
} # la.mix
if (li.mix) {
allx.i <- lowsup:max(i.mix)
pmf2.i <- dgaitdbinom(x, # Outer distribution---mlm type
size.i, prob.i,
truncate = setdiff(allx.i, i.mix))
}
sum.d <- 0 # numeric(LLL)
if (ld.mlm) {
pdip.mlm <- matrix(pdip.mlm, LLL, ld.mlm, byrow = byrow.aid)
sum.d <- .rowSums(pdip.mlm, LLL, ld.mlm)
if (any(1 < sum.d, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
} # ld.mlm
if (ld.mix) {
allx.d <- lowsup:max(d.mix)
pmf2.d <- dgaitdbinom(x, size.p = size.d, prob.p = prob.d,
truncate = setdiff(allx.d, d.mix))
} # ld.mix
sum.i <- 0
if (li.mlm) {
pstr.mlm <- matrix(pstr.mlm, LLL, li.mlm, byrow = byrow.aid)
sum.i <- .rowSums(pstr.mlm, LLL, li.mlm)
if (any(1 < sum.i, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
} # li.mlm
skip <- vecTF.t | vecTF.a # Leave these values alone
tmp6 <- 1 - sum.a - sum.i - pobs.mix - pstr.mix + sum.d + pdip.mix
if (li.mlm + ld.mlm) {
if (any(tmp6[!skip] < 0, na.rm = TRUE)) {
warning("the vector of normalizing constants contains ",
"some negative values. Replacing them with NAs")
tmp6[!skip & tmp6 < 0] <- NA
}
} # li.mlm + ld.mlm
pmf0[!skip] <- (tmp6 * # added
dbinom(x, size.p, prob.p) / (cdf.max.s - suma - sumt))[!skip]
if (li.mlm) {
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
if (any(vecTF <- is.finite(x) & ival == x)) {
pmf0[vecTF] <- pmf0[vecTF] + pstr.mlm[vecTF, jay]
}
} # jay
} # li.mlm
if (ld.mlm) {
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
if (any(vecTF <- is.finite(x) & dval == x)) {
pmf0[vecTF] <- pmf0[vecTF] - pdip.mlm[vecTF, jay]
}
} # jay
} # ld.mlm
pmf0 <- pmf0 + pobs.mix * pmf2.a + pstr.mix * pmf2.i -
pdip.mix * pmf2.d
if (log.arg) log(pmf0) else pmf0
} # dgaitdbinom
pgaitdbinom <-
function(q, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0,
pobs.mlm = 0,
pstr.mix = 0,
pstr.mlm = 0,
pdip.mix = 0,
pdip.mlm = 0,
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
lower.tail = TRUE,
...) { # ... is for max.support (ignored)
max.support <- Inf
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(pbinom(q, size.p, prob.p, lower.tail = lower.tail))
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(q),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a), length(size.i),
length(size.d),
length(prob.p), length(prob.a), length(prob.i),
length(prob.d))
offset.a <- offset.i <- offset.d <-
Offset.a <- Offset.i <- Offset.d <- numeric(LLL)
if (length(q) < LLL) q <- rep_len(q, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
sumt <- 0
fudge.t <- numeric(LLL)
cdf.max.s <- pbinom(max.support, size.p, prob.p) # Usually 1
if (ltrunc) {
for (tval in truncate) {
pmf.p <- dbinom(tval, size.p, prob.p)
sumt <- sumt + pmf.p
if (any(vecTF <- is.finite(q) & tval <= q))
fudge.t[vecTF] <- fudge.t[vecTF] + pmf.p[vecTF]
}
} # ltrunc
sum.a <- suma <- 0 # numeric(LLL)
fudge.a <- numeric(LLL)
if (la.mlm) {
pobs.mlm <- matrix(pobs.mlm, LLL, la.mlm, byrow = byrow.aid)
sum.a <- .rowSums(pobs.mlm, LLL, la.mlm)
if (any(1 < sum.a, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
for (jay in seq(la.mlm)) {
aval <- a.mlm[jay]
pmf.p <- dbinom(aval, size.p, prob.p)
suma <- suma + pmf.p # cumulative; part i
if (any(vecTF <- (is.finite(q) & aval <= q))) {
offset.a[vecTF] <- offset.a[vecTF] + pobs.mlm[vecTF, jay]
fudge.a[vecTF] <- fudge.a[vecTF] + pmf.p[vecTF] # cmtive
}
} # jay
} # la.mlm
sum.i <- 0
if (li.mlm) {
pstr.mlm <- matrix(pstr.mlm, LLL, li.mlm, byrow = byrow.aid)
sum.i <- .rowSums(pstr.mlm, LLL, li.mlm)
if (any(1 < sum.i, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
for (jay in seq(li.mlm)) {
ival <- i.mlm[jay]
if (any(vecTF <- (is.finite(q) & ival <= q))) {
offset.i[vecTF] <- offset.i[vecTF] + pstr.mlm[vecTF, jay]
}
} # jay
} # li.mlm
use.pobs.mix <- 0
if (la.mix) {
use.pobs.mix <- matrix(0, LLL, la.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
pmf.a <- dbinom(aval, size.a, prob.a)
pmf.p <- dbinom(aval, size.p, prob.p)
use.pobs.mix[, jay] <- pmf.a
suma <- suma + pmf.p # cumulative; part ii
}
use.pobs.mix <- pobs.mix *
use.pobs.mix / rowSums(use.pobs.mix)
for (jay in seq(la.mix)) {
aval <- a.mix[jay]
pmf.p <- dbinom(aval, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & aval <= q))) {
Offset.a[vecTF] <- Offset.a[vecTF] + use.pobs.mix[vecTF, jay]
fudge.a[vecTF] <- fudge.a[vecTF] + pmf.p[vecTF] # cumulative
}
} # jay
} # la.mix
use.pstr.mix <- 0
if (li.mix) {
use.pstr.mix <- matrix(0, LLL, li.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
use.pstr.mix[, jay] <- dbinom(ival, size.i, prob.i)
}
use.pstr.mix <- pstr.mix *
use.pstr.mix / rowSums(use.pstr.mix)
for (jay in seq(li.mix)) {
ival <- i.mix[jay]
pmf.p <- dbinom(ival, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & ival <= q))) {
Offset.i[vecTF] <- Offset.i[vecTF] + use.pstr.mix[vecTF, jay]
}
} # jay
} # li.mix
sum.d <- 0
if (ld.mlm) {
pdip.mlm <- matrix(pdip.mlm, LLL, ld.mlm, byrow = byrow.aid)
sum.d <- .rowSums(pdip.mlm, LLL, ld.mlm)
if (any(1 < sum.d, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
for (jay in seq(ld.mlm)) {
dval <- d.mlm[jay]
if (any(vecTF <- (is.finite(q) & dval <= q))) {
offset.d[vecTF] <- offset.d[vecTF] + pdip.mlm[vecTF, jay]
}
} # jay
} # ld.mlm
use.pdip.mix <- 0
if (ld.mix) {
use.pdip.mix <- matrix(0, LLL, ld.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
use.pdip.mix[, jay] <- dbinom(dval, size.d, prob.d)
}
use.pdip.mix <- pdip.mix *
use.pdip.mix / rowSums(use.pdip.mix)
for (jay in seq(ld.mix)) {
dval <- d.mix[jay]
pmf.p <- dbinom(dval, size.p, prob.p)
if (any(vecTF <- (is.finite(q) & dval <= q))) {
Offset.d[vecTF] <- Offset.d[vecTF] + use.pdip.mix[vecTF, jay]
}
} # jay
} # ld.mix
numer1 <- 1 - sum.i - sum.a - pstr.mix - pobs.mix +
sum.d + pdip.mix
denom1 <- cdf.max.s - sumt - suma
ans <- numer1 * (pbinom(q, size.p, prob.p) - fudge.t -
fudge.a) / denom1 +
offset.a + offset.i - offset.d +
Offset.a + Offset.i - Offset.d
ans[max.support <= q] <- 1
ans[ans < 0] <- 0 # Occasional roundoff error
if (lower.tail) ans else 1 - ans
} # pgaitdbinom
qgaitdbinom <-
function(p, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0,
pobs.mlm = 0,
pstr.mix = 0,
pstr.mlm = 0,
pdip.mix = 0,
pdip.mlm = 0,
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
...) { # ... is for max.support (ignored)
max.support <- NULL # Different from Inf
if (!length(max.support)) # Manually
max.support <- max(size.p, size.a, size.i, na.rm = TRUE)
lowsup <- 0
gaitd.errorcheck(a.mix, a.mlm, i.mix, i.mlm,
d.mix, d.mlm, truncate, max.support)
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm +
ltrunc == 0 &&
max.support >= max(size.p, na.rm = TRUE))
return(qbinom(p, size.p, prob.p )) # lower.tail, log.p = FALSE
if (la.mix == 0) pobs.mix <- 0
if (la.mlm == 0) pobs.mlm <- 0
if (li.mix == 0) pstr.mix <- 0
if (li.mlm == 0) pstr.mlm <- 0
if (ld.mix == 0) pdip.mix <- 0
if (ld.mlm == 0) pdip.mlm <- 0
if (any(pobs.mix < 0 | 1 <= pobs.mix, na.rm = TRUE))
stop("bad input for argument 'pobs.mix'")
if (any(pobs.mlm < 0 | 1 <= pobs.mlm, na.rm = TRUE))
stop("bad input for argument 'pobs.mlm'")
if (any(pstr.mix < 0 | 1 <= pstr.mix, na.rm = TRUE))
stop("bad input for argument 'pstr.mix'")
if (any(pstr.mlm < 0 | 1 <= pstr.mlm, na.rm = TRUE))
stop("bad input for argument 'pstr.mlm'")
if (any(pdip.mix < 0 | 1 <= pdip.mix, na.rm = TRUE))
stop("bad input for argument 'pdip.mix'")
if (any(pdip.mlm < 0 | 1 <= pdip.mlm, na.rm = TRUE))
stop("bad input for argument 'pdip.mlm'")
LLL <- max(length(p),
length(pobs.mix), length(pstr.mix), length(pdip.mix),
length(size.p), length(size.a),
length(size.i), length(size.d),
length(prob.p), length(prob.a),
length(prob.i), length(prob.d))
if (length(p) < LLL) p <- rep_len(p, LLL)
if (length(size.p) < LLL) size.p <- rep_len(size.p, LLL)
if (length(size.a) < LLL) size.a <- rep_len(size.a, LLL)
if (length(size.i) < LLL) size.i <- rep_len(size.i, LLL)
if (length(size.d) < LLL) size.d <- rep_len(size.d, LLL)
if (length(prob.p) < LLL) prob.p <- rep_len(prob.p, LLL)
if (length(prob.a) < LLL) prob.a <- rep_len(prob.a, LLL)
if (length(prob.i) < LLL) prob.i <- rep_len(prob.i, LLL)
if (length(prob.d) < LLL) prob.d <- rep_len(prob.d, LLL)
if (length(pobs.mix) < LLL) pobs.mix <- rep_len(pobs.mix, LLL)
if (length(pstr.mix) < LLL) pstr.mix <- rep_len(pstr.mix, LLL)
if (length(pdip.mix) < LLL) pdip.mix <- rep_len(pdip.mix, LLL)
pobs.mlm <- matrix(pobs.mlm, LLL, max(la.mlm, 1),
byrow = byrow.aid)
pstr.mlm <- matrix(pstr.mlm, LLL, max(li.mlm, 1),
byrow = byrow.aid)
pdip.mlm <- matrix(pdip.mlm, LLL, max(ld.mlm, 1),
byrow = byrow.aid)
min.support <- lowsup # Usual case; same as lowsup
min.support.use <- if (ltrunc)
min(setdiff(min.support:(ltrunc+5), truncate)) else
min.support
ans <- p + size.p + size.a + size.i + size.d +
prob.p + prob.a + prob.i + prob.d
bad0.p <- !is.finite(size.p) | size.p <= 0 |
!is.finite(prob.p) | prob.p <= 0 | 1 <= prob.p
bad0.a <- !is.finite(size.a) | size.a <= 0 |
!is.finite(prob.a) | prob.a <= 0 | 1 <= prob.a
bad0.i <- !is.finite(size.i) | size.i <= 0 |
!is.finite(prob.i) | prob.i <= 0 | 1 <= prob.i
bad0.d <- !is.finite(size.d) | size.d <= 0 |
!is.finite(prob.d) | prob.d <= 0 | 1 <= prob.d
bad0 <- bad0.p | bad0.a | bad0.i | bad0.d
bad <- bad0 | !is.finite(p) | p <= 0 | 1 <= p
Lo <- rep_len(min.support.use - 0.5, LLL)
approx.ans <- Lo # True at lhs
Hi <- rep_len(max.support + 0.5, LLL) # Need finite RHS endpoint
dont.iterate <- bad
done <- dont.iterate |
p <= pgaitdbinom(Hi, size.p, prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix, pobs.mix = pobs.mix,
pdip.mix = pdip.mix,
pstr.mlm = pstr.mlm, pobs.mlm = pobs.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i,
size.d = size.d,
prob.a = prob.a, prob.i = prob.i,
prob.d = prob.d,
byrow.aid = FALSE)
iter <- 0
max.iter <- round(log2(.Machine$double.xmax)) - 3
while (!all(done) && iter < max.iter) {
Lo[!done] <- Hi[!done]
Hi[!done] <- 2 * Hi[!done] + 10.5 # Bug fixed
Hi <- pmin(max.support + 0.5, Hi) # 20190924
done[!done] <-
(p[!done] <= pgaitdbinom(Hi[!done],
size.p[!done], prob.p[!done],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix[!done],
pstr.mix = pstr.mix[!done],
pdip.mix = pdip.mix[!done],
pobs.mlm = pobs.mlm[!done, , drop = FALSE],
pstr.mlm = pstr.mlm[!done, , drop = FALSE],
pdip.mlm = pdip.mlm[!done, , drop = FALSE],
size.a = size.a[!done],
size.i = size.i[!done],
size.d = size.d[!done],
prob.a = prob.a[!done],
prob.i = prob.i[!done],
prob.d = prob.d[!done],
byrow.aid = FALSE))
iter <- iter + 1
}
foo <- function(q, size.p, prob.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, pstr.mix = 0, pdip.mix = 0,
pobs.mlm = 0, pstr.mlm = 0, pdip.mlm = 0,
size.a = size.p, size.i = size.p,
size.d = size.p,
prob.a = prob.p, prob.i = prob.p,
prob.d = prob.p,
byrow.aid = FALSE, p)
pgaitdbinom(q, size.p = size.p, prob.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, prob.a = prob.a,
size.i = size.i, prob.i = prob.i,
size.d = size.d, prob.d = prob.d,
byrow.aid = FALSE) - p
lhs <- dont.iterate |
p <= dgaitdbinom(min.support.use,
size.p = size.p, prob.p = prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, prob.a = prob.a,
size.i = size.i, prob.i = prob.i,
size.d = size.d, prob.d = prob.d,
byrow.aid = FALSE)
if (any(!lhs)) {
approx.ans[!lhs] <-
bisection.basic(foo, Lo[!lhs], Hi[!lhs], tol = 1/16,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE,
p = p[!lhs])
faa <- floor(approx.ans[!lhs])
tmp <-
ifelse(pgaitdbinom(faa,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE) < p[!lhs] &
p[!lhs] <= pgaitdbinom(faa + 1,
size.p = size.p[!lhs],
prob.p = prob.p[!lhs],
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pstr.mix = pstr.mix[!lhs],
pstr.mlm = pstr.mlm[!lhs, , drop = FALSE],
pobs.mix = pobs.mix[!lhs],
pobs.mlm = pobs.mlm[!lhs, , drop = FALSE],
pdip.mix = pdip.mix[!lhs],
pdip.mlm = pdip.mlm[!lhs, , drop = FALSE],
size.a = size.a[!lhs],
prob.a = prob.a[!lhs],
size.i = size.i[!lhs],
prob.i = prob.i[!lhs],
size.d = size.d[!lhs],
prob.d = prob.d[!lhs],
byrow.aid = FALSE),
faa + 1, faa)
ans[!lhs] <- tmp
} # any(!lhs)
if (ltrunc)
while (any(vecTF <- !bad & ans %in% truncate))
ans[vecTF] <- 1 + ans[vecTF]
vecTF <- !bad0 & !is.na(p) &
p <= dgaitdbinom(min.support.use, size.p, prob.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pstr.mix = pstr.mix,
pdip.mix = pdip.mix,
pobs.mlm = pobs.mlm,
pstr.mlm = pstr.mlm,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i,
size.d = size.d,
prob.a = prob.a, prob.i = prob.i,
prob.d = prob.d,
byrow.aid = FALSE)
ans[vecTF] <- min.support.use
ans[!bad0 & !is.na(p) & p == 0] <- min.support.use
ans[!bad0 & !is.na(p) & p == 1] <- max.support # Inf
ans[!bad0 & !is.na(p) & p < 0] <- NaN
ans[!bad0 & !is.na(p) & p > 1] <- NaN
ans[ bad0] <- NaN
ans
} # qgaitdbinom
rgaitdbinom <-
function(n, size.p, prob.p,
a.mix = NULL,
a.mlm = NULL,
i.mix = NULL,
i.mlm = NULL,
d.mix = NULL,
d.mlm = NULL,
truncate = NULL,
pobs.mix = 0, # vector
pobs.mlm = 0, # matrix
pstr.mix = 0, # vector
pstr.mlm = 0, # matrix
pdip.mix = 0, # vector
pdip.mlm = 0, # matrix
byrow.aid = FALSE,
size.a = size.p, size.i = size.p, size.d = size.p,
prob.a = prob.p, prob.i = prob.p, prob.d = prob.p,
...) { # ... is for max.support (ignored)
qgaitdbinom(runif(n), size.p, prob.p,
a.mix = a.mix,
a.mlm = a.mlm,
i.mix = i.mix,
i.mlm = i.mlm,
d.mix = d.mix,
d.mlm = d.mlm,
truncate = truncate,
pobs.mix = pobs.mix,
pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix,
pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix,
pdip.mlm = pdip.mlm,
size.a = size.a, size.i = size.i, size.d = size.d,
prob.a = prob.a, prob.i = prob.i, prob.d = prob.d,
byrow.aid = byrow.aid)
} # rgaitdbinom
gaitd.errorcheck <-
function(a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL,
max.support = Inf,
min.support = 0, nparams = 1) {
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
ltrunc <- length(truncate)
if (!is.numeric(max.support) || is.na(max.support) ||
length(max.support) != 1 || max.support < min.support ||
round(max.support) != max.support ||
(length(truncate) && (
min(truncate, na.rm = TRUE) < min.support ||
max.support <= max(truncate, na.rm = TRUE))))
stop("bad input for argument 'max.support' and/or ",
"'truncate'")
allargs <- c(a.mix, a.mlm, i.mix, i.mlm, d.mix, d.mlm)
allargs <- c(allargs, truncate) # No NA, NaN, -Inf or Inf allowed
if (la.mix + la.mlm + li.mix + li.mlm + ld.mix + ld.mlm)
if (!is.Numeric(allargs, integer.valued = TRUE) ||
any(allargs < min.support) ||
any(max.support < allargs))
stop("bad input for arguments 'a.mix', 'a.mlm', ",
"'i.mix', 'i.mlm', 'd.mix' and/or 'd.mlm'")
if (length(unique(allargs)) < la.mix + la.mlm + li.mix + li.mlm +
ld.mix + ld.mlm + ltrunc)
stop("duplicate values found in arguments 'a.mix', ",
"'a.mlm', 'i.mix', 'i.mlm', 'd.mix', 'd.mlm'",
" and 'truncate'")
if (nparams == 2) {
if(la.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(a.mix) == ", la.mix,
" points")
if (li.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(i.mix) == ", li.mix,
" points")
if (ld.mix == 2)
stop("overfitting: trying to fit a ", nparams, "-parameter ",
"distribution based on length(d.mix) == ", ld.mix,
" points")
}
} # gaitd.errorcheck
moments.gaitdcombo.1par <-
function(theta.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, # Vector
pobs.mlm = 0, # Matrix
pstr.mix = 0,
pstr.mlm = 0, # Ditto
pdip.mix = 0,
pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
theta.a = theta.p, theta.i = theta.p, theta.d = theta.p,
moments2 = FALSE, # Use this for variances.
rmlife1 = 0, rmlife2 = 0,
dfun = "dpois") {
NOS <- 1
nnn <- length(theta.p)
pfun <- dfun
substring(pfun, 1) <- "p" # Replace the "d" by a "p"
cdf.max.s <- do.call(pfun, list(max.support, theta.p))
LALT.MIX <- length(a.mix)
LALT.MLM <- length(a.mlm)
LINF.MIX <- length(i.mix)
LINF.MLM <- length(i.mlm)
LDEF.MIX <- length(d.mix)
LDEF.MLM <- length(d.mlm)
LTRUNCAT <- length(truncate)
if (LALT.MLM == 0) {
if (!all(pobs.mlm == 0))
warning("ignoring argument 'pobs.mlm'")
pobs.mlm <- 0
}
if (LINF.MLM == 0) {
if (!all(pstr.mlm == 0))
warning("ignoring argument 'pstr.mlm'")
pstr.mlm <- 0
}
if (LDEF.MLM == 0) {
if (!all(pdip.mlm == 0))
warning("ignoring argument 'pdip.mlm'")
pdip.mlm <- 0
}
if (LALT.MIX == 0) {
if (!all(pobs.mix == 0))
warning("ignoring argument 'pobs.mix'")
pobs.mix <- 0
}
if (LINF.MIX == 0) {
if (!all(pstr.mix == 0))
warning("ignoring argument 'pstr.mix'")
pstr.mix <- 0
}
if (LDEF.MIX == 0) {
if (!all(pdip.mix == 0))
warning("ignoring argument 'pdip.mix'")
pdip.mix <- 0
}
SumT0.p <- matrix(0, nnn, NOS) # Does not include upper RHS tail
SumT1.p <- matrix(rmlife1, nnn, NOS) # Includes RHS tail
SumT2.p <- matrix(rmlife2, nnn, NOS) # Includes RHS tail
if (LTRUNCAT)
for (tval in truncate) {
pmf.p <- do.call(dfun, list(tval, theta.p))
SumT0.p <- SumT0.p + pmf.p # Need tval<=max.support
SumT1.p <- SumT1.p + pmf.p * tval
if (moments2)
SumT2.p <- SumT2.p + pmf.p * tval^2
}
use.pobs.mix <- use.pobs.mlm <- # So rowSums() works below.
use.pstr.mix <- use.pstr.mlm <-
use.pdip.mix <- use.pdip.mlm <- matrix(0, nnn, 1)
aprd1.mix <- aprd1.mlm <- # aprd1.m?? is an innerprod
aprd2.mix <- aprd2.mlm <- 0 # aprd2.m?? is an innerprod
SumA0.mix.p <- SumA0.mlm.p <-
SumA0.mix.a <- SumA0.mlm.a <-
SumA1.mix.p <- SumA1.mlm.p <-
SumA1.mix.a <- SumA1.mlm.a <-
SumA1.mlm.x <-
SumA2.mix.p <- SumA2.mlm.p <-
SumA2.mix.a <- SumA2.mlm.a <-
SumA2.mlm.x <- matrix(0, nnn, NOS)
if (LALT.MIX)
use.pobs.mix <- matrix(pobs.mix, nnn, 1)
if (LINF.MIX)
use.pstr.mix <- matrix(pstr.mix, nnn, 1)
if (LDEF.MIX)
use.pdip.mix <- matrix(pdip.mix, nnn, 1)
if (LALT.MLM)
use.pobs.mlm <- matrix(pobs.mlm, nnn, LALT.MLM,
byrow = byrow.aid)
if (LINF.MLM)
use.pstr.mlm <- matrix(pstr.mlm, nnn, LINF.MLM,
byrow = byrow.aid)
if (LDEF.MLM)
use.pdip.mlm <- matrix(pdip.mlm, nnn, LDEF.MLM,
byrow = byrow.aid)
if (LALT.MIX) {
for (jay in seq_len(LALT.MIX)) {
aval <- a.mix[jay]
pmf.p <- do.call(dfun, list(aval, theta.p))
pmf.a <- do.call(dfun, list(aval, theta.a))
SumA0.mix.p <- SumA0.mix.p + pmf.p
SumA0.mix.a <- SumA0.mix.a + pmf.a
SumA1.mix.p <- SumA1.mix.p + pmf.p * aval
SumA1.mix.a <- SumA1.mix.a + pmf.a * aval
if (moments2) {
SumA2.mix.p <- SumA2.mix.p + pmf.p * aval^2
SumA2.mix.a <- SumA2.mix.a + pmf.a * aval^2
}
} # for jay
aprd1.mix <- use.pobs.mix * SumA1.mix.a / SumA0.mix.a
if (moments2)
aprd2.mix <- use.pobs.mix * SumA2.mix.a / SumA0.mix.a
} # LALT.MIX
if (LALT.MLM) {
for (jay in seq_len(LALT.MLM)) {
aval <- a.mlm[jay]
pmf.x <- use.pobs.mlm[, jay]
pmf.p <- do.call(dfun, list(aval, theta.p))
pmf.a <- do.call(dfun, list(aval, theta.a))
SumA0.mlm.p <- SumA0.mlm.p + pmf.p
SumA0.mlm.a <- SumA0.mlm.a + pmf.a
SumA1.mlm.p <- SumA1.mlm.p + pmf.p * aval
SumA1.mlm.a <- SumA1.mlm.a + pmf.a * aval
SumA1.mlm.x <- SumA1.mlm.x + pmf.x * aval
if (moments2) {
SumA2.mlm.p <- SumA2.mlm.p + pmf.p * aval^2
SumA2.mlm.a <- SumA2.mlm.a + pmf.a * aval^2
SumA2.mlm.x <- SumA2.mlm.x + pmf.x * aval^2
}
} # for jay
aprd1.mlm <- SumA1.mlm.x
if (moments2)
aprd2.mlm <- SumA2.mlm.x
} # LALT.MLM
iprd1.mix <- iprd1.mlm <- # iprd1.m?? is an innerprod
iprd2.mix <- iprd2.mlm <- 0 # iprd2.m?? is an innerprod
SumI0.mix.p <- SumI0.mlm.p <-
SumI0.mix.i <- SumI0.mlm.i <-
SumI1.mix.p <- SumI1.mlm.p <-
SumI1.mix.i <- SumI1.mlm.i <-
SumI1.mlm.x <-
SumI2.mix.p <- SumI2.mlm.p <-
SumI2.mix.i <- SumI2.mlm.i <-
SumI2.mlm.x <- matrix(0, nnn, NOS)
dprd1.mix <- dprd1.mlm <-
dprd2.mix <- dprd2.mlm <- 0
SumD0.mix.p <- SumD0.mlm.p <-
SumD0.mix.d <- SumD0.mlm.d <-
SumD1.mix.p <- SumD1.mlm.p <-
SumD1.mix.d <- SumD1.mlm.d <-
SumD1.mlm.x <-
SumD2.mix.p <- SumD2.mlm.p <-
SumD2.mix.d <- SumD2.mlm.d <-
SumD2.mlm.x <- matrix(0, nnn, NOS)
if (LINF.MIX) {
for (jay in seq_len(LINF.MIX)) {
ival <- i.mix[jay]
pmf.p <- do.call(dfun, list(ival, theta.p))
pmf.i <- do.call(dfun, list(ival, theta.i))
SumI0.mix.p <- SumI0.mix.p + pmf.p
SumI0.mix.i <- SumI0.mix.i + pmf.i
SumI1.mix.p <- SumI1.mix.p + pmf.p * ival
SumI1.mix.i <- SumI1.mix.i + pmf.i * ival
if (moments2) {
SumI2.mix.p <- SumI2.mix.p + pmf.p * ival^2
SumI2.mix.i <- SumI2.mix.i + pmf.i * ival^2
}
} # for jay
iprd1.mix <- use.pstr.mix * SumI1.mix.i / SumI0.mix.i
if (moments2)
iprd2.mix <- use.pstr.mix * SumI2.mix.i / SumI0.mix.i
} # LINF.MIX
if (LINF.MLM) {
for (jay in seq_len(LINF.MLM)) {
ival <- i.mlm[jay]
pmf.x <- use.pstr.mlm[, jay]
pmf.p <- do.call(dfun, list(ival, theta.p))
pmf.i <- do.call(dfun, list(ival, theta.i))
SumI0.mlm.p <- SumI0.mlm.p + pmf.p
SumI0.mlm.i <- SumI0.mlm.i + pmf.i
SumI1.mlm.p <- SumI1.mlm.p + pmf.p * ival
SumI1.mlm.i <- SumI1.mlm.i + pmf.i * ival
SumI1.mlm.x <- SumI1.mlm.x + pmf.x * ival
if (moments2) {
SumI2.mlm.p <- SumI2.mlm.p + pmf.p * ival^2
SumI2.mlm.i <- SumI2.mlm.i + pmf.i * ival^2
SumI2.mlm.x <- SumI2.mlm.x + pmf.x * ival^2
}
} # for jay
iprd1.mlm <- SumI1.mlm.x
if (moments2)
iprd2.mlm <- SumI2.mlm.x
} # LINF.MLM
if (LDEF.MIX) {
for (jay in seq_len(LDEF.MIX)) {
dval <- d.mix[jay]
pmf.p <- do.call(dfun, list(dval, theta.p))
pmf.d <- do.call(dfun, list(dval, theta.d))
SumD0.mix.p <- SumD0.mix.p + pmf.p
SumD0.mix.d <- SumD0.mix.d + pmf.d
SumD1.mix.p <- SumD1.mix.p + pmf.p * dval
SumD1.mix.d <- SumD1.mix.d + pmf.d * dval
if (moments2) {
SumD2.mix.p <- SumD2.mix.p + pmf.p * dval^2
SumD2.mix.d <- SumD2.mix.d + pmf.d * dval^2
}
} # for jay
dprd1.mix <- use.pdip.mix * SumD1.mix.d / SumD0.mix.d
if (moments2)
dprd2.mix <- use.pdip.mix * SumD2.mix.d / SumD0.mix.d
} # LDEF.MIX
if (LDEF.MLM) {
for (jay in seq_len(LDEF.MLM)) {
dval <- d.mlm[jay]
pmf.x <- use.pdip.mlm[, jay]
pmf.p <- do.call(dfun, list(dval, theta.p))
pmf.d <- do.call(dfun, list(dval, theta.d))
SumD0.mlm.p <- SumD0.mlm.p + pmf.p
SumD0.mlm.d <- SumD0.mlm.d + pmf.d
SumD1.mlm.p <- SumD1.mlm.p + pmf.p * dval
SumD1.mlm.d <- SumD1.mlm.d + pmf.d * dval
SumD1.mlm.x <- SumD1.mlm.x + pmf.x * dval
if (moments2) {
SumD2.mlm.p <- SumD2.mlm.p + pmf.p * dval^2
SumD2.mlm.d <- SumD2.mlm.d + pmf.d * dval^2
SumD2.mlm.x <- SumD2.mlm.x + pmf.x * dval^2
}
} # for jay
dprd1.mlm <- SumD1.mlm.x
if (moments2)
dprd2.mlm <- SumD2.mlm.x
} # LDEF.MLM
use.this <- 1 - rowSums(use.pobs.mlm) - rowSums(use.pstr.mlm) +
rowSums(use.pdip.mlm) -
use.pobs.mix - use.pstr.mix + use.pdip.mix
ans <- list('cdf.max.s' = cdf.max.s,
'SumT0.p' = SumT0.p,
'SumT1.p' = SumT1.p,
'SumA0.mix.a' = SumA0.mix.a,
'SumA0.mix.p' = SumA0.mix.p,
'SumA1.mix.a' = SumA1.mix.a,
'SumA1.mix.p' = SumA1.mix.p,
'SumA0.mlm.a' = SumA0.mlm.a,
'SumA0.mlm.p' = SumA0.mlm.p,
'SumA1.mlm.a' = SumA1.mlm.a,
'SumA1.mlm.p' = SumA1.mlm.p,
'SumI0.mix.i' = SumI0.mix.i,
'SumI0.mix.p' = SumI0.mix.p,
'SumI1.mix.i' = SumI1.mix.i,
'SumI1.mix.p' = SumI1.mix.p,
'SumI0.mlm.i' = SumI0.mlm.i,
'SumI0.mlm.p' = SumI0.mlm.p,
'SumI1.mlm.i' = SumI1.mlm.i,
'SumI1.mlm.p' = SumI1.mlm.p,
'SumD0.mix.d' = SumD0.mix.d,
'SumD0.mix.p' = SumD0.mix.p,
'SumD1.mix.d' = SumD1.mix.d,
'SumD1.mix.p' = SumD1.mix.p,
'SumD0.mlm.d' = SumD0.mlm.d,
'SumD0.mlm.p' = SumD0.mlm.p,
'SumD1.mlm.d' = SumD1.mlm.d,
'SumD1.mlm.p' = SumD1.mlm.p,
'aprd1.mix' = aprd1.mix,
'aprd1.mlm' = aprd1.mlm,
'iprd1.mix' = iprd1.mix,
'iprd1.mlm' = iprd1.mlm,
'dprd1.mix' = dprd1.mix, #
'dprd1.mlm' = dprd1.mlm, #
'use.this' = use.this)
if (moments2) { # Add more info
ans <- c(ans,
list( # 'rmlife2' = rmlife2, # May be scalar
'aprd2.mix' = aprd2.mix,
'aprd2.mlm' = aprd2.mlm,
'iprd2.mix' = iprd2.mix,
'iprd2.mlm' = iprd2.mlm,
'dprd2.mix' = dprd2.mix, #
'dprd2.mlm' = dprd2.mlm, #
'SumT2.p' = SumT2.p,
'SumA2.mix.p' = SumA2.mix.p,
'SumA2.mix.a' = SumA2.mix.a,
'SumI2.mix.p' = SumI2.mix.p,
'SumI2.mix.i' = SumI2.mix.i,
'SumD2.mix.p' = SumD2.mix.p, #
'SumD2.mix.d' = SumD2.mix.d, #
'SumA2.mlm.p' = SumA2.mlm.p,
'SumA2.mlm.a' = SumA2.mlm.a,
'SumI2.mlm.p' = SumI2.mlm.p,
'SumI2.mlm.i' = SumI2.mlm.i,
'SumD2.mlm.p' = SumD2.mlm.p, #
'SumD2.mlm.d' = SumD2.mlm.d)) #
}
ans
} # moments.gaitdcombo.1par
moments.gaitdcombo.pois <-
function(lambda.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, # Vector
pobs.mlm = 0, # Matrix
pstr.mix = 0,
pstr.mlm = 0, # Ditto
pdip.mix = 0,
pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
lambda.a = lambda.p, lambda.i = lambda.p,
lambda.d = lambda.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
rmlife1 <- ppois(max.support - 1, lambda.p, lower.tail = FALSE) *
lambda.p
rmlife2 <- ppois(max.support - 2, lambda.p, lower.tail = FALSE) *
lambda.p^2 + rmlife1
mylist1 <- moments.gaitdcombo.1par(theta.p = lambda.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = lambda.a, theta.i = lambda.i,
theta.d = lambda.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dpois")
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this * (lambda.p - SumA1.mix.p -
SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.pois
moments.gaitdcombo.log <-
function(shape.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
shape.a = shape.p, shape.i = shape.p, shape.d = shape.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
A8.p <- -1 / log1p(-shape.p)
rmlife1 <- A8.p * (shape.p^(max.support + 1)) / (1 - shape.p)
rmlife2 <- A8.p * ((shape.p^(max.support + 1)) *
(max.support + 1 / (1 - shape.p))
/ (1 - shape.p))
mylist1 <- moments.gaitdcombo.1par(theta.p = shape.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = shape.a, theta.i = shape.i, theta.d = shape.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dlog")
themean <- with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this *
(-shape.p / (log1p(-shape.p) * (1 - shape.p)) -
SumA1.mix.p - SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.log
moments.gaitdcombo.zeta <-
function(shape.p,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
truncate = NULL, max.support = Inf,
pobs.mix = 0, pobs.mlm = 0, # Vector and matrix resp.
pstr.mix = 0, pstr.mlm = 0, # Ditto
pdip.mix = 0, pdip.mlm = 0, # Ditto
byrow.aid = FALSE, # For pobs.mlm and pstr.mlm
shape.a = shape.p, shape.i = shape.p, shape.d = shape.p,
type.fitted = "All", # or "mean"
moments2 = FALSE) { # Use this for variances.
rmlife1 <- if (is.finite(max.support)) zeta(shape.p) * (1 -
pzeta(max.support, shape.p - 1)) / zeta(shape.p + 1) else
numeric(length(shape.p))
rmlife1[shape.p <= 1] <- NA # NA or Inf, not sure
rmlife2 <- if (is.finite(max.support)) zeta(shape.p - 1) * (1 -
pzeta(max.support, shape.p - 2)) / zeta(shape.p + 1) else
numeric(length(shape.p))
rmlife2[shape.p <= 2] <- NA # NA or Inf, not sure
mylist1 <- moments.gaitdcombo.1par(theta.p = shape.p,
a.mix = a.mix, a.mlm = a.mlm,
i.mix = i.mix, i.mlm = i.mlm,
d.mix = d.mix, d.mlm = d.mlm,
truncate = truncate, max.support = max.support,
pobs.mix = pobs.mix, pobs.mlm = pobs.mlm,
pstr.mix = pstr.mix, pstr.mlm = pstr.mlm,
pdip.mix = pdip.mix, pdip.mlm = pdip.mlm,
byrow.aid = byrow.aid, # type.fitted = type.fitted,
theta.a = shape.a, theta.i = shape.i, theta.d = shape.d,
moments2 = moments2,
rmlife1 = rmlife1, rmlife2 = rmlife2,
dfun = "dzeta")
themean <-
with(mylist1,
aprd1.mix + iprd1.mix + aprd1.mlm + iprd1.mlm -
dprd1.mix - dprd1.mlm +
use.this *
(ifelse(shape.p > 1, zeta(shape.p) / zeta(shape.p + 1), NA) -
SumA1.mix.p - SumA1.mlm.p - SumT1.p) / (
cdf.max.s - SumA0.mix.p - SumA0.mlm.p - SumT0.p))
if (type.fitted == "mean") {
return(themean)
}
ans <- c(mylist1,
list('rmlife1' = rmlife1, # Has the right dimension
'mean' = themean))
if (moments2) { # Add more info
ans <- c(ans,
list('rmlife2' = rmlife2))
}
ans
} # moments.gaitdcombo.zeta
specialsvglm <-
function(object, ...) {
infos <- object@family@infos()
ans <- list(a.mix = infos$a.mix,
a.mlm = infos$a.mlm,
i.mix = infos$i.mix,
i.mlm = infos$i.mlm,
d.mix = infos$d.mix,
d.mlm = infos$d.mlm,
truncate = infos$truncate)
if (is.numeric(tmp7e <- infos$max.support))
ans <- c(ans, max.support = tmp7e)
ans
} # specialsvglm
if (!isGeneric("specials"))
setGeneric("specials", function(object, ...)
standardGeneric("specials"),
package = "VGAM")
setMethod("specials", "vglm",
function(object, ...)
specialsvglm(object, ...))
if (!isGeneric("altered"))
setGeneric("altered", function(object, ...)
standardGeneric("altered"),
package = "VGAM")
setMethod("altered", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
c(tmp$a.mix, tmp$a.mlm)})
if (!isGeneric("inflated"))
setGeneric("inflated", function(object, ...)
standardGeneric("inflated"),
package = "VGAM")
setMethod("inflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
c(tmp$i.mix, tmp$i.mlm)})
if (!isGeneric("truncated"))
setGeneric("truncated", function(object, ...)
standardGeneric("truncated"),
package = "VGAM")
setMethod("truncated", "vglm",
function(object, ...) {
ans <- specialsvglm(object, ...)
if (any(names(ans) == "max.support"))
ans[c("truncate", "max.support")] else
ans[["truncate"]]
})
setGeneric("is.altered", function(object, ...)
standardGeneric("is.altered"),
package = "VGAM")
setMethod("is.altered", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$a.mix, tmp$a.mlm)))})
setGeneric("is.inflated", function(object, ...)
standardGeneric("is.inflated"),
package = "VGAM")
setMethod("is.inflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$i.mix, tmp$i.mlm)))})
setGeneric("is.deflated", function(object, ...)
standardGeneric("is.deflated"),
package = "VGAM")
setMethod("is.deflated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(c(tmp$d.mix, tmp$d.mlm)))})
setGeneric("is.truncated", function(object, ...)
standardGeneric("is.truncated"),
package = "VGAM")
setMethod("is.truncated", "vglm",
function(object, ...) {
tmp <- specialsvglm(object, ...)
as.logical(length(tmp$truncated)) ||
(length(tmp$max.support) > 0 && is.finite(tmp$max.support))
})
y.gaitcombo.check <-
function(y, truncate = NULL,
a.mix = NULL, a.mlm = NULL,
i.mix = NULL, i.mlm = NULL,
d.mix = NULL, d.mlm = NULL,
max.support = Inf, min.support = 0) {
la.mix <- length(a.mix <- sort(a.mix))
li.mix <- length(i.mix <- sort(i.mix))
ld.mix <- length(d.mix <- sort(d.mix))
la.mlm <- length(a.mlm)
li.mlm <- length(i.mlm)
ld.mlm <- length(d.mlm)
n <- length(y)
css.mix.a <- css.mix.i <- css.mix.d <-
css.mlm.a <- css.mlm.i <- css.mlm.d <- NULL
skip.mix.a <- skip.mix.i <- skip.mix.d <- # Default
skip.mlm.a <- skip.mlm.i <- skip.mlm.d <- NULL
if (length(truncate) && any(y %in% truncate))
stop("some response values == values in argument 'truncate'")
if (max.support < max(y))
stop("some response values are greater than the ",
"'max.support' argument")
y0.mix.a <- y0.mlm.a <-
y0.mix.i <- y0.mlm.i <-
y0.mix.d <- y0.mlm.d <- NULL
if (la.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.a <- matrix(0, n, la.mix)
for (jay in seq(la.mix))
y0.mix.a[, jay] <- as.numeric(y == a.mix[jay])
skip.mix.a <- matrix(as.logical(y0.mix.a), n, la.mix) # dim lost
if (any((css.mix.a <- colSums(skip.mix.a)) == 0))
stop("some 'a.mix' argument values have no response values: ",
paste(a.mix[css.mix.a == 0], collapse = ", "))
} # la.mix
if (la.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.a <- matrix(0, n, la.mlm)
for (jay in seq(la.mlm))
y0.mlm.a[, jay] <- as.numeric(y == a.mlm[jay])
skip.mlm.a <- matrix(as.logical(y0.mlm.a), n, la.mlm) # dim lost
if (any((css.mlm.a <- colSums(skip.mlm.a)) == 0))
stop("some 'a.mlm' argument values have no response values: ",
paste(a.mlm[css.mlm.a == 0], collapse = ", "))
} # la.mlm
if (li.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.i <- matrix(0, n, li.mix)
for (jay in seq(li.mix))
y0.mix.i[, jay] <- as.numeric(y == i.mix[jay])
skip.mix.i <- matrix(as.logical(y0.mix.i), n, li.mix) # dim lost
if (any((css.mix.i <- colSums(skip.mix.i)) == 0))
stop("some 'i.mix' argument values have no response values: ",
paste(i.mix[css.mix.i == 0], collapse = ", "))
} # li.mix
if (li.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.i <- matrix(0, n, li.mlm)
for (jay in seq(li.mlm))
y0.mlm.i[, jay] <- as.numeric(y == i.mlm[jay])
skip.mlm.i <- matrix(as.logical(y0.mlm.i), n, li.mlm) # dim lost
if (any((css.mlm.i <- colSums(skip.mlm.i)) == 0))
stop("some 'i.mlm' argument values have no response values: ",
paste(i.mlm[css.mlm.i == 0], collapse = ", "))
} # li.mlm
if (ld.mix > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mix.d <- matrix(0, n, ld.mix)
for (jay in seq(ld.mix))
y0.mix.d[, jay] <- as.numeric(y == d.mix[jay])
skip.mix.d <- matrix(as.logical(y0.mix.d), n, ld.mix) # dim lost
if (any((css.mix.d <- colSums(skip.mix.d)) == 0))
stop("some 'd.mix' argument values have no response values: ",
paste(d.mix[css.mix.d == 0], collapse = ", "))
} # ld.mix
if (ld.mlm > 0) { # ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
y0.mlm.d <- matrix(0, n, ld.mlm)
for (jay in seq(ld.mlm))
y0.mlm.d[, jay] <- as.numeric(y == d.mlm[jay])
skip.mlm.d <- matrix(as.logical(y0.mlm.d), n, ld.mlm) # dim lost
if (any((css.mlm.d <- colSums(skip.mlm.d)) == 0))
stop("some 'd.mlm' argument values have no response values: ",
paste(d.mlm[css.mlm.d == 0], collapse = ", "))
} # ld.mlm
list(css.mix.a = css.mix.a, skip.mix.a = skip.mix.a,
css.mix.i = css.mix.i, skip.mix.i = skip.mix.i,
css.mix.d = css.mix.d, skip.mix.d = skip.mix.d,
css.mlm.a = css.mlm.a, skip.mlm.a = skip.mlm.a,
css.mlm.i = css.mlm.i, skip.mlm.i = skip.mlm.i,
css.mlm.d = css.mlm.d, skip.mlm.d = skip.mlm.d,
y0.mix.a = y0.mix.a, y0.mlm.a = y0.mlm.a,
y0.mix.i = y0.mix.i, y0.mlm.i = y0.mlm.i,
y0.mix.d = y0.mix.d, y0.mlm.d = y0.mlm.d)
} # y.gaitcombo.check
Try the VGAM package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.