Nothing
R/regression_models.R
In Rfast: A Collection of Efficient and Extremely Fast R Functions
Defines functions
spml.reg
spatmed.reg
poisson.cat1
normlog.reg
multinom.reg
logistic.cat1
lmfit
gammareg
bc
ar1
Documented in
ar1
bc
gammareg
lmfit
logistic.cat1
multinom.reg
normlog.reg
poisson.cat1
spatmed.reg
spml.reg
#[export]
ar1 <- function(y, method = "cmle") {
N <- length(y)
if ( method == "cmle" ) {
dera2 <- N - 1
derab <- sum( y[-N] )
derb2 <- sum( y[-N]^2 )
dera <- derab
derb <- sum( y[-N] * y[-1])
cphi <- c( derb2 * dera - derab * derb, - derab * dera + dera2 * derb ) / ( dera2 * derb2 - derab^2 )
s <- sum( (y[-1] - cphi[1] - cphi[2] * y[-N])^2 )/dera2
param <- c(cphi, s)
names(param) <- c("constant", "phi", "sigma")
} else if ( method == "yw" ) {
m <- sum(y)/N
z <- y - m
phi <- sum( z[-N] * z[-1] ) / sum(z^2)
sigma <- (1 - phi^2) * sum(z^2)/(N - 2)
param <- c(m, phi, sigma)
names(param) <- c("mean", "phi", "sigma")
} else if ( method == "ols" ) {
m <- sum(y)/N
z <- y - m
y1 <- z[-1]
x1 <- z[-N]
phi <- as.vector( cov(y1, x1) )/ Var(x1)
constant <- mean(y1) - phi * mean(x1)
sigma <- sum( (y1 - constant - phi * x1 )^2 ) / (N - 1)
param <- c(constant, phi, sigma)
names(param) <- c("constant", "phi", "sigma")
}
param
}
#[export]
bc <- function(x, low = -1, up = 1) {
lx <- log(x)
slx <- sum( lx )
n2 <- - 0.5 * length(x)
vlx <- Rfast::Var(lx)
ell <- function(lambda, x, vlx, slx, n2) {
if ( abs(lambda) < 1e-12 ) {
s <- vlx
} else {
y <- x^lambda
s <- Rfast::Var(y) / lambda^2
}
n2 * log(s) + lambda * slx
}
optimise(ell, c(low, up), x = x, vlx = vlx, slx = slx, n2 = n2,
maximum = TRUE, tol = 1e-06)$maximum
}
#[export]
gammareg <- function(y, x, tol = 1e-07, maxiters = 100) {
X <- model.matrix( y~., data.frame(x) )
sx <- Rfast::colsums(X)
dm <- dim(X)
n <- dm[1] ; p <- dm[2]
mod <- Rfast::gammacon(y, tol = tol)
m <- exp(mod$be)
be <- c( mod$be, numeric(p - 1) )
d1 <- - sum(y / m) + n * mod$be
con <- y / m
der <- Rfast::eachcol.apply(X, con) - sx
der2 <- - crossprod( con * X, X)
be <- be - solve(der2, der)
xbe <- X %*% be
m <- as.vector( exp( xbe ) )
con <- y / m
d2 <- - sum(con) - sum(xbe)
i <- 2
while ( abs(d1 - d2) > tol & i < maxiters) {
i <- i + 1
d1 <- d2
der <- Rfast::eachcol.apply(X, con) - sx
der2 <- - crossprod( con * X, X)
be <- be - solve(der2, der)
xbe <- X %*% be
m <- as.vector( exp( xbe ) )
con <- y / m
d2 <- - sum(con) - sum(xbe)
}
phi <- sum( (con - 1)^2 ) / (n - p)
devi <- - 2 * sum( log(con) ) + sum(con) - n
info <- c(i, devi, phi)
names(info) <- c("iters", "deviance", "phi")
names(be) <- colnames(X)
list(info = info, be = be)
}
#[export]
glm_logistic <- function (x, y, full = FALSE, tol = 1e-09,maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_glm_logistic, x, y,tol,maxiters)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance)
if (full) {
be <- mod$be
se <- chol2inv(chol(mod$der2))
se <- sqrt(diag(se))
wald <- be/se
pval <- 2 * pnorm(abs(wald), lower.tail = FALSE)
info <- cbind(be, se, wald, pval)
colnames(info) <- c("estimate", "std error", "Wald stat",
"p-value")
rownames(info) <- colnames(x)
res <- list(info = info, devi = mod$deviance)
}
res["iter"] <- mod$iter
res
}
#[export]
glm_poisson <- function (x, y, full = FALSE,tol = 1e-09) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_glm_poisson, x, y, sum(y * log(y), na.rm = TRUE),tol)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance)
if (full) {
be <- mod$be
se <- chol2inv(chol(mod$L2))
se <- sqrt(diag(se))
wald <- be/se
pval <- 2 * pnorm(abs(wald), lower.tail = FALSE)
info <- cbind(be, se, wald, pval)
colnames(info) <- c("estimate", "std error", "Wald stat",
"p-value")
rownames(info) <- colnames(x)
res <- list(info = info, devi = mod$deviance)
}
res
}
#[export]
invgauss.reg <- function (y, x, tol = 1e-07, maxiters = 100) {
X <- model.matrix(~., data.frame(x))
dm <- dim(X)
ly <- log(y)
y2 <- 2 * y
n <- dm[1]
sy <- sum(1/y)
be1 <- solve( crossprod(X), crossprod(X, ly) )
mi <- exp(as.vector(X %*% be1))
f <- Rfast::colsums((mi - y)/mi^2 * X)
f2 <- crossprod((y2/mi^2 - 1/mi) * X, X)
be2 <- be1 - solve(f2, f)
i <- 2
while ( sum(abs(be1 - be2)) > tol & i < maxiters ) {
i <- i + 1
be1 <- be2
mi <- as.vector(exp(X %*% be1))
f <- Rfast::colsums((mi - y)/mi^2 * X)
f2 <- crossprod((y2/mi^2 - 1/mi) * X, X)
be2 <- be1 - solve(f2, f)
}
lambda <- 1/(sy/n - sum(1/mi)/n)
loglik <- n/2 * log(lambda/2/pi) - 1.5 * sum(ly) - lambda/2 *
sum( (-y/mi^2 + sy/n) )
deviance <- n/lambda
phi <- sum((y - mi)^2/(mi^3))/(n - 2)
names(be2) <- colnames(X)
list(iters = i, loglik = loglik, deviance = deviance, phi = phi,
be = be2)
}
#[export]
lmfit <- function(x, y, w = NULL) {
if ( is.null(w) ) {
be <- solve( crossprod(x), crossprod(x, y) )
} else be <- solve( crossprod(x, w * x), crossprod(x, w * y) )
e <- as.vector( y - x %*% be )
rownames(be) <- colnames(x)
list(be = be, residuals = e)
}
#[export]
logistic.cat1 <- function(y, x, logged = FALSE) {
N <- Rfast::Table(y, x)
cj <- log( N[2, ] / N[1, ] )
be <- c(cj[1], cj[-1] - cj[1])
n <- length(y)
Nj <- Rfast::colsums( N )
pj <- N[2, ] / Nj
p <- sum(N[2, ]) / n
D0 <- - 2 * n * sum( p * log(p) + (1 - p) * log(1 - p) )
D1 <- - 2 * sum( Nj * pj * log(pj) + Nj * (1 - pj) * log(1 - pj) )
se <- pj * (1 - pj) * Nj
se <- c( 1/se[1], 1 / se[-1] + 1/se[1] )
se <- sqrt(se)
stat <- be/se
pval <- pchisq(stat^2, 1, lower.tail = FALSE)
mat <- cbind(be, se, stat^2, pval)
colnames(mat) <- c("estimate", "std. error", "Wald", "p-value")
devs <- c(D0, D1)
stat <- D0 - D1
pvalue <- pchisq(stat, length(be) - 1, lower.tail = FALSE, log.p = logged)
devs <- c(D0, D1, stat, pvalue)
names(devs) <- c("null deviance", "residual deviance", "stat", "p-value")
list(info = mat, devs = devs)
}
#[export]
multinom.reg <- function(y, x, tol = 1e-07, maxiters = 50) {
Y <- Rfast::design_matrix(y)[, -1]
if ( !is.matrix(Y) ) {
mod <- Rfast::glm_logistic(x, y, tol = tol, maxiters = maxiters)
res <- list(iters = mod$iter, loglik = 0.5 * mod$devi, be = mod$be)
} else {
X <- model.matrix(y~., data.frame(x))
p <- dim(X)[2]
dm <- dim(Y)
n <- dm[1]
d <- dm[2]
m <- Rfast::colmeans(Y)
b0 <- Rfast::Log(m / (1 - m) )
b1 <- matrix( c(b0, numeric(p * d - d) ), nrow = p, ncol = d, byrow = TRUE)
e <- Y - rep(m, rep(n, d) )
id <- matrix(1:c(p * d), ncol = d)
der <- numeric(d * p)
der2 <- matrix(0, p * d, p * d)
for (i in 1:d) {
der[id[, i]] <- Rfast::colsums( e[, i] * X )
for (j in i:d) {
if (i != j) {
der2[id[, i], id[, j]] <- der2[id[, j], id[, i]] <- - crossprod(m[i] * m[j] * X, X)
} else der2[id[, i], id[, i]] <- crossprod(m[i] * (1 - m[i]) * X, X)
}
}
b2 <- b1 + solve(der2, der)
k <- 2
res <- try(while (sum(abs(b2 - b1)) > tol & k < maxiters) {
k <- k + 1
b1 <- b2
m1 <- exp(X %*% b1)
m <- m1 / (rowsums(m1) + 1)
e <- Y - m
for (i in 1:d) {
der[id[, i]] <- Rfast::colsums( e[, i] * X )
for (j in i:d) {
if (i != j) {
der2[id[, i], id[, j]] <- der2[id[, j], id[, i]] <- - crossprod(m[, i] * m[, j] * X, X)
} else der2[id[, i], id[, i]] <- crossprod(m[, i] * (1 - m[, i]) * X, X)
}
}
b2 <- b1 + solve(der2, der)
}, silent = TRUE)
if (inherits(res,"try-error")) b2 <- b1
colnames(b2) <- paste("Y", 1:d, sep = "")
rownames(b2) <- colnames(X)
Y1 <- design_matrix(y, ones = FALSE)
m <- cbind(1, m1)
m <- m / rowsums(m)
loglik <- - sum(Y1 * log(Y1/m), na.rm = TRUE )
res <- list(iters = k, loglik = loglik, be = b2)
}
res
}
#[export]
normlog.reg <- function(y, x, tol = 1e-07, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_normlog_reg,y, x, tol, maxiters)
rownames(mod$be) <- colnames(x)
mod
}
#[export]
poisson.cat1 <- function(y, x, logged = FALSE) {
ni <- tabulate(x)
si <- rowsum(y, x)
mi <- si/ni
n <- length(y)
k <- length(ni)
be <- log(mi)
be <- c(be[1], be[-1] - be[1])
s <- matrix(0, k, k)
diag(s) <- si
s[1, ] <- si
s[, 1] <- si
s[1, 1] <- sum(si)
se <- sqrt( diag( chol2inv(chol(s)) ) )
#########
cl1 <- sum( si * log(mi) )
cl0 <- s[1, 1] * log( s[1, 1]/n )
stat <- 2 * cl1 - 2 * cl0
pvalue <- pchisq(stat, k - 1, lower.tail = FALSE, log.p = logged)
res <- c(stat, pvalue)
names(res) <- c("stat", "p-value")
########
stat <- be^2/se^2
pval <- pchisq(stat, 1, lower.tail = FALSE)
mat <- cbind(be, se, stat, pval)
colnames(mat) <- c("estimate", "std. error", "Wald", "p-value")
list(info = mat, res = res)
}
#[export]
prop.reg <- function (y, x, varb = "quasi", tol = 1e-09, maxiters = 100) {
X <- model.matrix(~., data.frame(x))
L <- .Call(Rfast_prop_reg, X, y, tol, maxiters)
der2 <- L$der2
u <- y - as.vector(L$p)
be <- as.vector(L$be)
names(be) <- colnames(X)
Ainv <- spdinv(der2)
phi <- NULL
if (varb == "quasi") {
B <- crossprod(u * X)
vb <- Ainv %*% B %*% Ainv
}
else {
dm <- dim(X)
dof <- dm[1] - dm[2]
phi <- sum((y - L$p)^2/L$p/(1 - L$p))/dof
vb <- phi * Ainv
}
info <- cbind(be, sqrt(diag(vb)), be^2/diag(vb))
info <- cbind(info, pchisq(info[, 3], 1, lower.tail = FALSE))
rownames(info) <- colnames(X)
colnames(info) <- c("Estimate", "Std. error", "Wald", "p-value")
list(iters = L$i, varb = vb, phi = phi, info = info)
}
################################
#### Spatial median regression
#### Tsagris Michail 10/2014
#### Biman Chakraborty (2003) On multivariate quantile regression
#### Journal of Statistical Planning and Inference
#### http://www.stat.nus.edu.sg/export/sites/dsap/research/documents/tr01_2000.pdf
################################
#[export]
spatmed.reg <- function(y, x, tol = 1e-07) {
x <- model.matrix(y ~ ., data.frame(x) )
p <- dim(x)[2]
d <- dim(y)[2]
B1 <- solve(crossprod(x), crossprod(x, y) )
est <- y - x %*% B1
ww <- sqrt( Rfast::rowsums(est^2) )
## ww <- sqrt( Rfast::rowsums( est^2 ) )
z <- x / ww
B2 <- solve( crossprod(z, x), crossprod(z, y) )
i <- 2
while ( sum( abs(B2 - B1) ) > tol ) {
i <- i + 1
B1 <- B2
est <- y - x %*% B1
ww <- sqrt( Rfast::rowsums(est^2) )
## ww <- sqrt( Rfast::rowsums( est^2 ) )
ela <- which( ww == 0 )
z <- x / ww
if ( length(ela) > 0 ) z[ela, ] <- 0
B2 <- solve( crossprod(z, x), crossprod(z, y) )
}
be <- B2
if ( is.null(colnames(y)) ) {
colnames(be) <- paste("Y", 1:d, sep = "")
} else colnames(be) <- colnames(y)
rownames(be) <- colnames(x)
list(iters = i, be = be)
}
#[export]
spml.reg <- function(y, x, tol = 1e-07, seb = FALSE, maxiters = 100) {
x <- model.matrix(~., data.frame(x) )
y <- as.matrix(y)
mod <- .Call(Rfast_spml_reg, y, x, tol, seb, maxiters)
colnames(mod$be) <- c("Cosinus of y", "Sinus of y")
rownames(mod$be) <- colnames(x)
if ( seb ) {
colnames(mod$seb) <- c("Cosinus of y", "Sinus of y")
rownames(mod$seb) <- colnames(x)
} else mod$seb <- NULL
mod
}
#[export]
weib.reg <- function (y, x, tol = 1e-07, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_weib_reg, y, x, tol,
maxiters)
rownames(mod$be) <- colnames(x)
list(iters = mod$iters, loglik = mod$loglik, shape = mod$shape, be = mod$be)
}
#[export]
qpois.reg <- function (x, y, full = FALSE, tol = 1e-09, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_qpois_reg, x, y, sum(y * log(y), na.rm = TRUE),
tol,maxiters)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance, varb = mod$phi *
spdinv(mod$L2), phi = mod$phi)
if (full) {
be <- mod$be
varb <- mod$phi * spdinv(mod$L2)
info <- cbind(be, sqrt(diag(varb)), be^2/diag(varb))
info <- cbind(info, pchisq(info[, 3], 1, lower.tail = FALSE))
rownames(info) <- colnames(x)
colnames(info) <- c("Estimate", "Std. error", "Wald",
"p-value")
res <- list(info = info, devi = mod$deviance, varb = varb,
phi = mod$phi)
}
res
}
Try the Rfast package in your browser
Any scripts or data that you put into this service are public.
Rfast documentation built on Feb. 16, 2023, 5:09 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions spml.reg spatmed.reg poisson.cat1 normlog.reg multinom.reg logistic.cat1 lmfit gammareg bc ar1
Documented in ar1 bc gammareg lmfit logistic.cat1 multinom.reg normlog.reg poisson.cat1 spatmed.reg spml.reg
#[export]
ar1 <- function(y, method = "cmle") {
N <- length(y)
if ( method == "cmle" ) {
dera2 <- N - 1
derab <- sum( y[-N] )
derb2 <- sum( y[-N]^2 )
dera <- derab
derb <- sum( y[-N] * y[-1])
cphi <- c( derb2 * dera - derab * derb, - derab * dera + dera2 * derb ) / ( dera2 * derb2 - derab^2 )
s <- sum( (y[-1] - cphi[1] - cphi[2] * y[-N])^2 )/dera2
param <- c(cphi, s)
names(param) <- c("constant", "phi", "sigma")
} else if ( method == "yw" ) {
m <- sum(y)/N
z <- y - m
phi <- sum( z[-N] * z[-1] ) / sum(z^2)
sigma <- (1 - phi^2) * sum(z^2)/(N - 2)
param <- c(m, phi, sigma)
names(param) <- c("mean", "phi", "sigma")
} else if ( method == "ols" ) {
m <- sum(y)/N
z <- y - m
y1 <- z[-1]
x1 <- z[-N]
phi <- as.vector( cov(y1, x1) )/ Var(x1)
constant <- mean(y1) - phi * mean(x1)
sigma <- sum( (y1 - constant - phi * x1 )^2 ) / (N - 1)
param <- c(constant, phi, sigma)
names(param) <- c("constant", "phi", "sigma")
}
param
}
#[export]
bc <- function(x, low = -1, up = 1) {
lx <- log(x)
slx <- sum( lx )
n2 <- - 0.5 * length(x)
vlx <- Rfast::Var(lx)
ell <- function(lambda, x, vlx, slx, n2) {
if ( abs(lambda) < 1e-12 ) {
s <- vlx
} else {
y <- x^lambda
s <- Rfast::Var(y) / lambda^2
}
n2 * log(s) + lambda * slx
}
optimise(ell, c(low, up), x = x, vlx = vlx, slx = slx, n2 = n2,
maximum = TRUE, tol = 1e-06)$maximum
}
#[export]
gammareg <- function(y, x, tol = 1e-07, maxiters = 100) {
X <- model.matrix( y~., data.frame(x) )
sx <- Rfast::colsums(X)
dm <- dim(X)
n <- dm[1] ; p <- dm[2]
mod <- Rfast::gammacon(y, tol = tol)
m <- exp(mod$be)
be <- c( mod$be, numeric(p - 1) )
d1 <- - sum(y / m) + n * mod$be
con <- y / m
der <- Rfast::eachcol.apply(X, con) - sx
der2 <- - crossprod( con * X, X)
be <- be - solve(der2, der)
xbe <- X %*% be
m <- as.vector( exp( xbe ) )
con <- y / m
d2 <- - sum(con) - sum(xbe)
i <- 2
while ( abs(d1 - d2) > tol & i < maxiters) {
i <- i + 1
d1 <- d2
der <- Rfast::eachcol.apply(X, con) - sx
der2 <- - crossprod( con * X, X)
be <- be - solve(der2, der)
xbe <- X %*% be
m <- as.vector( exp( xbe ) )
con <- y / m
d2 <- - sum(con) - sum(xbe)
}
phi <- sum( (con - 1)^2 ) / (n - p)
devi <- - 2 * sum( log(con) ) + sum(con) - n
info <- c(i, devi, phi)
names(info) <- c("iters", "deviance", "phi")
names(be) <- colnames(X)
list(info = info, be = be)
}
#[export]
glm_logistic <- function (x, y, full = FALSE, tol = 1e-09,maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_glm_logistic, x, y,tol,maxiters)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance)
if (full) {
be <- mod$be
se <- chol2inv(chol(mod$der2))
se <- sqrt(diag(se))
wald <- be/se
pval <- 2 * pnorm(abs(wald), lower.tail = FALSE)
info <- cbind(be, se, wald, pval)
colnames(info) <- c("estimate", "std error", "Wald stat",
"p-value")
rownames(info) <- colnames(x)
res <- list(info = info, devi = mod$deviance)
}
res["iter"] <- mod$iter
res
}
#[export]
glm_poisson <- function (x, y, full = FALSE,tol = 1e-09) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_glm_poisson, x, y, sum(y * log(y), na.rm = TRUE),tol)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance)
if (full) {
be <- mod$be
se <- chol2inv(chol(mod$L2))
se <- sqrt(diag(se))
wald <- be/se
pval <- 2 * pnorm(abs(wald), lower.tail = FALSE)
info <- cbind(be, se, wald, pval)
colnames(info) <- c("estimate", "std error", "Wald stat",
"p-value")
rownames(info) <- colnames(x)
res <- list(info = info, devi = mod$deviance)
}
res
}
#[export]
invgauss.reg <- function (y, x, tol = 1e-07, maxiters = 100) {
X <- model.matrix(~., data.frame(x))
dm <- dim(X)
ly <- log(y)
y2 <- 2 * y
n <- dm[1]
sy <- sum(1/y)
be1 <- solve( crossprod(X), crossprod(X, ly) )
mi <- exp(as.vector(X %*% be1))
f <- Rfast::colsums((mi - y)/mi^2 * X)
f2 <- crossprod((y2/mi^2 - 1/mi) * X, X)
be2 <- be1 - solve(f2, f)
i <- 2
while ( sum(abs(be1 - be2)) > tol & i < maxiters ) {
i <- i + 1
be1 <- be2
mi <- as.vector(exp(X %*% be1))
f <- Rfast::colsums((mi - y)/mi^2 * X)
f2 <- crossprod((y2/mi^2 - 1/mi) * X, X)
be2 <- be1 - solve(f2, f)
}
lambda <- 1/(sy/n - sum(1/mi)/n)
loglik <- n/2 * log(lambda/2/pi) - 1.5 * sum(ly) - lambda/2 *
sum( (-y/mi^2 + sy/n) )
deviance <- n/lambda
phi <- sum((y - mi)^2/(mi^3))/(n - 2)
names(be2) <- colnames(X)
list(iters = i, loglik = loglik, deviance = deviance, phi = phi,
be = be2)
}
#[export]
lmfit <- function(x, y, w = NULL) {
if ( is.null(w) ) {
be <- solve( crossprod(x), crossprod(x, y) )
} else be <- solve( crossprod(x, w * x), crossprod(x, w * y) )
e <- as.vector( y - x %*% be )
rownames(be) <- colnames(x)
list(be = be, residuals = e)
}
#[export]
logistic.cat1 <- function(y, x, logged = FALSE) {
N <- Rfast::Table(y, x)
cj <- log( N[2, ] / N[1, ] )
be <- c(cj[1], cj[-1] - cj[1])
n <- length(y)
Nj <- Rfast::colsums( N )
pj <- N[2, ] / Nj
p <- sum(N[2, ]) / n
D0 <- - 2 * n * sum( p * log(p) + (1 - p) * log(1 - p) )
D1 <- - 2 * sum( Nj * pj * log(pj) + Nj * (1 - pj) * log(1 - pj) )
se <- pj * (1 - pj) * Nj
se <- c( 1/se[1], 1 / se[-1] + 1/se[1] )
se <- sqrt(se)
stat <- be/se
pval <- pchisq(stat^2, 1, lower.tail = FALSE)
mat <- cbind(be, se, stat^2, pval)
colnames(mat) <- c("estimate", "std. error", "Wald", "p-value")
devs <- c(D0, D1)
stat <- D0 - D1
pvalue <- pchisq(stat, length(be) - 1, lower.tail = FALSE, log.p = logged)
devs <- c(D0, D1, stat, pvalue)
names(devs) <- c("null deviance", "residual deviance", "stat", "p-value")
list(info = mat, devs = devs)
}
#[export]
multinom.reg <- function(y, x, tol = 1e-07, maxiters = 50) {
Y <- Rfast::design_matrix(y)[, -1]
if ( !is.matrix(Y) ) {
mod <- Rfast::glm_logistic(x, y, tol = tol, maxiters = maxiters)
res <- list(iters = mod$iter, loglik = 0.5 * mod$devi, be = mod$be)
} else {
X <- model.matrix(y~., data.frame(x))
p <- dim(X)[2]
dm <- dim(Y)
n <- dm[1]
d <- dm[2]
m <- Rfast::colmeans(Y)
b0 <- Rfast::Log(m / (1 - m) )
b1 <- matrix( c(b0, numeric(p * d - d) ), nrow = p, ncol = d, byrow = TRUE)
e <- Y - rep(m, rep(n, d) )
id <- matrix(1:c(p * d), ncol = d)
der <- numeric(d * p)
der2 <- matrix(0, p * d, p * d)
for (i in 1:d) {
der[id[, i]] <- Rfast::colsums( e[, i] * X )
for (j in i:d) {
if (i != j) {
der2[id[, i], id[, j]] <- der2[id[, j], id[, i]] <- - crossprod(m[i] * m[j] * X, X)
} else der2[id[, i], id[, i]] <- crossprod(m[i] * (1 - m[i]) * X, X)
}
}
b2 <- b1 + solve(der2, der)
k <- 2
res <- try(while (sum(abs(b2 - b1)) > tol & k < maxiters) {
k <- k + 1
b1 <- b2
m1 <- exp(X %*% b1)
m <- m1 / (rowsums(m1) + 1)
e <- Y - m
for (i in 1:d) {
der[id[, i]] <- Rfast::colsums( e[, i] * X )
for (j in i:d) {
if (i != j) {
der2[id[, i], id[, j]] <- der2[id[, j], id[, i]] <- - crossprod(m[, i] * m[, j] * X, X)
} else der2[id[, i], id[, i]] <- crossprod(m[, i] * (1 - m[, i]) * X, X)
}
}
b2 <- b1 + solve(der2, der)
}, silent = TRUE)
if (inherits(res,"try-error")) b2 <- b1
colnames(b2) <- paste("Y", 1:d, sep = "")
rownames(b2) <- colnames(X)
Y1 <- design_matrix(y, ones = FALSE)
m <- cbind(1, m1)
m <- m / rowsums(m)
loglik <- - sum(Y1 * log(Y1/m), na.rm = TRUE )
res <- list(iters = k, loglik = loglik, be = b2)
}
res
}
#[export]
normlog.reg <- function(y, x, tol = 1e-07, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_normlog_reg,y, x, tol, maxiters)
rownames(mod$be) <- colnames(x)
mod
}
#[export]
poisson.cat1 <- function(y, x, logged = FALSE) {
ni <- tabulate(x)
si <- rowsum(y, x)
mi <- si/ni
n <- length(y)
k <- length(ni)
be <- log(mi)
be <- c(be[1], be[-1] - be[1])
s <- matrix(0, k, k)
diag(s) <- si
s[1, ] <- si
s[, 1] <- si
s[1, 1] <- sum(si)
se <- sqrt( diag( chol2inv(chol(s)) ) )
#########
cl1 <- sum( si * log(mi) )
cl0 <- s[1, 1] * log( s[1, 1]/n )
stat <- 2 * cl1 - 2 * cl0
pvalue <- pchisq(stat, k - 1, lower.tail = FALSE, log.p = logged)
res <- c(stat, pvalue)
names(res) <- c("stat", "p-value")
########
stat <- be^2/se^2
pval <- pchisq(stat, 1, lower.tail = FALSE)
mat <- cbind(be, se, stat, pval)
colnames(mat) <- c("estimate", "std. error", "Wald", "p-value")
list(info = mat, res = res)
}
#[export]
prop.reg <- function (y, x, varb = "quasi", tol = 1e-09, maxiters = 100) {
X <- model.matrix(~., data.frame(x))
L <- .Call(Rfast_prop_reg, X, y, tol, maxiters)
der2 <- L$der2
u <- y - as.vector(L$p)
be <- as.vector(L$be)
names(be) <- colnames(X)
Ainv <- spdinv(der2)
phi <- NULL
if (varb == "quasi") {
B <- crossprod(u * X)
vb <- Ainv %*% B %*% Ainv
}
else {
dm <- dim(X)
dof <- dm[1] - dm[2]
phi <- sum((y - L$p)^2/L$p/(1 - L$p))/dof
vb <- phi * Ainv
}
info <- cbind(be, sqrt(diag(vb)), be^2/diag(vb))
info <- cbind(info, pchisq(info[, 3], 1, lower.tail = FALSE))
rownames(info) <- colnames(X)
colnames(info) <- c("Estimate", "Std. error", "Wald", "p-value")
list(iters = L$i, varb = vb, phi = phi, info = info)
}
################################
#### Spatial median regression
#### Tsagris Michail 10/2014
#### Biman Chakraborty (2003) On multivariate quantile regression
#### Journal of Statistical Planning and Inference
#### http://www.stat.nus.edu.sg/export/sites/dsap/research/documents/tr01_2000.pdf
################################
#[export]
spatmed.reg <- function(y, x, tol = 1e-07) {
x <- model.matrix(y ~ ., data.frame(x) )
p <- dim(x)[2]
d <- dim(y)[2]
B1 <- solve(crossprod(x), crossprod(x, y) )
est <- y - x %*% B1
ww <- sqrt( Rfast::rowsums(est^2) )
## ww <- sqrt( Rfast::rowsums( est^2 ) )
z <- x / ww
B2 <- solve( crossprod(z, x), crossprod(z, y) )
i <- 2
while ( sum( abs(B2 - B1) ) > tol ) {
i <- i + 1
B1 <- B2
est <- y - x %*% B1
ww <- sqrt( Rfast::rowsums(est^2) )
## ww <- sqrt( Rfast::rowsums( est^2 ) )
ela <- which( ww == 0 )
z <- x / ww
if ( length(ela) > 0 ) z[ela, ] <- 0
B2 <- solve( crossprod(z, x), crossprod(z, y) )
}
be <- B2
if ( is.null(colnames(y)) ) {
colnames(be) <- paste("Y", 1:d, sep = "")
} else colnames(be) <- colnames(y)
rownames(be) <- colnames(x)
list(iters = i, be = be)
}
#[export]
spml.reg <- function(y, x, tol = 1e-07, seb = FALSE, maxiters = 100) {
x <- model.matrix(~., data.frame(x) )
y <- as.matrix(y)
mod <- .Call(Rfast_spml_reg, y, x, tol, seb, maxiters)
colnames(mod$be) <- c("Cosinus of y", "Sinus of y")
rownames(mod$be) <- colnames(x)
if ( seb ) {
colnames(mod$seb) <- c("Cosinus of y", "Sinus of y")
rownames(mod$seb) <- colnames(x)
} else mod$seb <- NULL
mod
}
#[export]
weib.reg <- function (y, x, tol = 1e-07, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_weib_reg, y, x, tol,
maxiters)
rownames(mod$be) <- colnames(x)
list(iters = mod$iters, loglik = mod$loglik, shape = mod$shape, be = mod$be)
}
#[export]
qpois.reg <- function (x, y, full = FALSE, tol = 1e-09, maxiters = 100) {
x <- model.matrix(y ~ ., data.frame(x))
mod <- .Call(Rfast_qpois_reg, x, y, sum(y * log(y), na.rm = TRUE),
tol,maxiters)
rownames(mod$be) <- colnames(x)
res <- list(be = mod$be, devi = mod$deviance, varb = mod$phi *
spdinv(mod$L2), phi = mod$phi)
if (full) {
be <- mod$be
varb <- mod$phi * spdinv(mod$L2)
info <- cbind(be, sqrt(diag(varb)), be^2/diag(varb))
info <- cbind(info, pchisq(info[, 3], 1, lower.tail = FALSE))
rownames(info) <- colnames(x)
colnames(info) <- c("Estimate", "Std. error", "Wald",
"p-value")
res <- list(info = info, devi = mod$deviance, varb = varb,
phi = mod$phi)
}
res
}
Try the Rfast 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.