Nothing
R/tibber.R
In distillery: Method Functions for Confidence Intervals and to Distill Information from an Object
Defines functions
plot.tibRMed
print.tibRMed
plot.tibbed
print.tibbed
tibberRM
tibber
Documented in
plot.tibbed
plot.tibRMed
print.tibbed
print.tibRMed
tibber
tibberRM
tibber <- function( x, statistic, B, rmodel, test.pars, rsize, block.length = 1, v.terms,
shuffle = NULL, replace = TRUE, alpha = 0.05, verbose = FALSE, ... ) {
begin.tiid <- Sys.time()
theCall <- match.call()
out <- list()
#
# 'rmodel' is a random generator function that minimally takes arguments:
# 'data', 'par', 'n', and '...', where 'par' is a single parameter value
# 'n' is the sample size to draw.
#
if( missing( x ) ) stop( "tibber: Must supply x argument." )
if( missing( statistic ) ) stop( "tibber: Must supply statistic argument." )
if( missing( B ) ) stop( "tibber: Must supply B argument." )
if( missing( rmodel ) ) stop( "tibber: Must supply rmodel argument." )
if( missing( test.pars ) ) stop( "tibber: test.pars argument not supplied." )
nT <- length( test.pars )
if( missing( rsize ) ) {
if( is.null( dim( x ) ) ) n <- length( x )
else n <- dim( x )[ 1 ]
rsize <- n
} else n <- rsize
obs <- do.call( statistic, c( list( data = x ), list( ... ) ) )
if( !missing( v.terms ) ) {
if( length( obs ) > 2 ) stop( "tibber: statistic must return only one value, or one value and its variance." )
obs.v <- obs[ v.terms ]
obs <- obs[ -v.terms ]
} else if( length( obs ) > 1 ) stop( "tibber: statistic must return only one value." )
# end of 'v.terms' missing or not stmts.
tfun <- function( tp, x, st, B, rmodel, N, rsize, bl, v.terms,
shuffle, replace, o, osd, ... ) {
xstar <- do.call( rmodel, c( list( data = x, par = tp, n = N ), list( ... ) ) )
if( verbose ) cat( "Nuisance parameter = ", tp, "\n" )
if( missing( rsize ) && missing( v.terms ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, block.length = bl,
shuffle = shuffle, replace = replace, ... )
} else if( missing( rsize ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, block.length = bl,
v.terms = v.terms, shuffle = shuffle, replace = replace, ... )
} else if( missing( v.terms ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, rsize = rsize, block.length = bl,
shuffle = shuffle, replace = replace, ... )
} else {
bootobj <- booter( x = xstar, statistic = st, B = B, rsize = rsize, block.length = bl,
v.terms = v.terms, shuffle = shuffle, replace = replace, ... )
}
# Qk <- do.call( st, c( list( data = xstar[ c( bootobj$indices ) ], list( ... ) ) ) )
Qk <- bootobj$original.est
if( !missing( v.terms ) ) Qk.v <- bootobj$orig.v
# TIB is not defined for more than one parameter at a time, but for now allowing
# for the possibility.
if( is.null( dim( bootobj$results ) ) ) {
B2 <- sum( !is.na( bootobj$results ), na.rm = TRUE )
Qmk <- quantile( bootobj$results, probs = 0.5 )
Pk <- ( sum( bootobj$results < o, na.rm = TRUE ) + sum( bootobj$results == o, na.rm = TRUE ) / 2 ) / B2
res <- c( Qk, Qmk, Pk )
if( !missing( v.terms ) ) {
Qmk.sd <- sqrt( bootobj$v )
Tstar <- ( bootobj$results - Qk ) / Qmk.sd
ot <- ( o - Qk ) / osd
Pk.stud <- ( sum( Tstar < ot, na.rm = TRUE ) + sum( Tstar == ot, na.rm = TRUE ) / 2 ) / B2
# Qk.stud <- quantile( Tstar, probs = 0.5 )
res <- c( res, Pk.stud )
} # end of if do studentized intervals too stmt.
} else {
B2 <- colSums( !is.na( bootobj$results ), na.rm = TRUE )
Qmk <- apply( bootobj$results, 2, quantile, probs = 0.5 )
bigO <- matrix( o, nrow = nrow( bootobj$results ), ncol = B )
o1 <- bootobj$results < bigO
o2 <- bootobj$results == bigO
Pk <- rowSums( o1 + o2 / 2, na.rm = TRUE ) / B2
res <- c( Qk, Qmk, Pk )
if( !missing( v.terms ) ) {
Qmk.sd <- matrix( sqrt( bootobj$v ), nrow = nrow( bootobj$results ), ncol = B )
Qkmat <- matrix( Qk, nrow = nrow( bootobj$results ), ncol = B )
Tstar <- ( bootobj$results - Qkmat ) / Qmk.sd
ot <- matrix( ( o - Qk ) / osd, nrow = nrow( bootobj$results ), ncol = B )
o1 <- Tstar < ot
o2 <- Tstar == ot
Pk.stud <- ( rowSums( o1 + o2 / 2, na.rm = TRUE ) ) / B2
# Qk.stud <- apply( Tstar, 1, quantile, probs = 0.5 )
res <- c( res, Pk.stud )
} # end of if do studentized intervals too stmt.
}
return( res )
} # end of internal 'tfun' function.
if( missing( rsize ) && missing( v.terms ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B,
rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, ... )
} else if( missing( rsize ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B,
v.terms = v.terms, rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, osd = sqrt( obs.v ), ... )
} else if( missing( v.terms ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B, rsize = rsize,
rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, ... )
} else {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B, rsize = rsize,
v.terms = v.terms, rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, osd = sqrt( obs.v ), ... )
}
if( missing( v.terms ) ) rownames( res ) <- c( "Est. Parameter", "Bootstrap Median Est. Parameter", "Est. P-value" )
else rownames( res ) <- c( "Est. Parameter", "Bootstrap Median Est. Parameter", "Est. P-value", "Est. P-value (Stud.)" )
colnames( res ) <- test.pars
out$results <- res
if( nT > 1 ) { # Do interpolation method.
# 'res' is a 3 (or 4 if v.terms present) by k matrix.
P <- res[ 3, ]
Q <- res[ 1, ]
a2 <- alpha / 2
q1 <- min( which( P < 1 - a2 ), na.rm = TRUE )
q2 <- max( which( P >= 1 - a2 ), na.rm = TRUE )
P1 <- P[ q1 ]
P2 <- P[ q2 ]
Q1 <- Q[ q1 ]
Q2 <- Q[ q2 ]
L <- ( Q1 - Q2 ) * ( P2 - ( 1 - a2 ) ) / ( P2 - P1 ) + Q2
out$Plow <- c( P1, P2 )
# Plow.est <- P1 + ( P2 - P1 ) * ( L - Q1 ) / ( Q2 - Q1 )
q1 <- min( which( P <= a2 ), na.rm = TRUE )
q2 <- max( which( P > a2 ), na.rm = TRUE )
P1 <- P[ q1 ]
P2 <- P[ q2 ]
Q1 <- Q[ q1 ]
Q2 <- Q[ q2 ]
U <- ( Q1 - Q2 ) * ( P2 - a2 ) / ( P2 - P1 ) + Q2
out$Pup <- c( P1, P2 )
# Note: if it is possible to have multiple statistics, then the following will need modification.
tib <- c( L, obs, U )
names( tib ) <- c( "Lower", "Estimate", "Upper" )
# Pup.est <- P1 + ( P2 - P1 ) * ( U - Q1 ) / ( Q2 - Q1 )
# pval.estimated <- c( 1 - Plow.est, 1 - Pup.est )
# names( pval.estimated ) <- c( "lower", "upper" )
out$TIB.interpolated <- tib
if( !missing( v.terms ) ) {
Pstud <- res[ 4, ]
# Qstud <- res[ 5, ]
q1stud <- min( which( Pstud < 1 - a2 ), na.rm = TRUE )
q2stud <- max( which( Pstud >= 1 - a2 ), na.rm = TRUE )
P1stud <- Pstud[ q1stud ]
P2stud <- Pstud[ q2stud ]
Q1 <- Q[ q1stud ]
Q2 <- Q[ q2stud ]
Lstud <- ( Q1 - Q2 ) * ( P2stud - ( 1 - a2 ) ) / ( P2stud - P1stud ) + Q2
out$PstudLow <- c( P1stud, P2stud )
# PstudLow <- P1stud + ( P2stud - P1stud ) * ( Lstud - Q1 ) / ( Q2 - Q1 )
q1stud <- min( which( Pstud <= a2 ), na.rm = TRUE )
q2stud <- max( which( Pstud > a2 ), na.rm = TRUE )
P1stud <- Pstud[ q1stud ]
P2stud <- Pstud[ q2stud ]
Q1 <- Q[ q1stud ]
Q2 <- Q[ q2stud ]
Ustud <- ( Q1 - Q2 ) * ( P2stud - a2 ) / ( P2stud - P1stud ) + Q2
stib <- c( Lstud, obs, Ustud )
names( stib ) <- c( "Lower", "Estimate", "Upper" )
out$STIB.interpolated <- stib
out$PstudUp <- c( P1stud, P2stud )
# PstudUp <- P1stud + ( P2stud - P1stud ) * ( Ustud - Q1 ) / (Q2 - Q1 )
# pval.estimated <- rbind( pval.estimated, c( 1 - PstudLow, 1 - PstudUp ) )
# rownames( pval.estimated ) <- c( "regular", "studentized" )
} # end of if do STIB interval stmt.
# out$estimated.pvalue <- pval.estimated
out$call <- theCall
out$data <- x
out$statistic <- statistic
out$B <- B
out$rmodel <- rmodel
out$n <- n
out$test.pars <- test.pars
out$rsize <- rsize
out$block.length <- block.length
if( !missing( v.terms ) ) out$v.terms <- v.terms
out$shuffle <- shuffle
out$replace <- replace
out$alpha <- alpha
tot.tiid <- Sys.time() - begin.tiid
out$total.time <- tot.tiid
if( verbose ) print( tot.tiid )
class( out ) <- "tibbed"
return( out )
} else return( res )
} # end of 'tibber' function.
tibberRM <- function( x, statistic, B, rmodel, startval, rsize, block.length = 1, v.terms,
shuffle = NULL, replace = TRUE, alpha = 0.05, step.size, tol = 1e-4, max.iter = 1000,
keep.iters = TRUE, verbose = FALSE, ... ) {
if( missing( startval ) ) stop( "tibberRM: must specify startval." )
begin.tiid <- Sys.time()
theCall <- match.call()
out <- list()
out$call <- theCall
out$x <- x
out$statistic <- statistic
out$B <- B
out$rmodel <- rmodel
if( !missing( v.terms ) ) out$v.terms <- v.terms
out$shuffle <- shuffle
out$alpha <- alpha
out$step.size <- step.size
out$tolerance <- tol
out$extra.args <- list( ... )
out$block.length <- block.length
out$replace <- replace
if( missing( rsize ) ) {
if( is.null( dim( x ) ) ) n <- length( x )
else n <- dim( x )[ 1 ]
rsize <- n
} else n <- rsize
out$rsize <- rsize
# Specify starting values for the nuisance parameter.
if( length( startval ) == 1 ) Lstart <- Ustart <- startval
else if( length( startval ) == 2 ) {
Lstart <- startval[ 1 ]
Ustart <- startval[ 2 ]
} else stop( "tibberRM: invalid startval argument." )
a2 <- alpha / 2
if( verbose ) {
cat( "\nBeginning to find lower bound.\n" )
cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", Lstart, "\n" )
} # end of if 'verbose' stmt.
# Do lower end-point.
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Lstart, rsize = rsize, block.length = block.length,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Lstart, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} # end of find starting value for parameter of lower end-point.
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
dP <- abs( pval.star - a2 )
th.star <- Lstart
if( keep.iters ) lower.iters <- c( par.star, th.star, pval.star, dP )
i <- 0
if( verbose ) cat( "Initial estimate (p-value) = ", par.star, "(", pval.star, ")\n" )
while( ( dP > tol ) && ( i < max.iter ) ) {
i <- i + 1
th.n <- th.star - step.size * ( ( 1 - a2 ) - pval.star ) / i
# if( verbose ) cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", th.n, "\n" )
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
}
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
th.star <- th.n
if( keep.iters ) lower.iters <- rbind( lower.iters, c( par.star, th.star, pval.star, abs( pval.star - ( 1 - a2 ) ) ) )
dP <- abs( pval.star - ( 1 - a2 ) )
} # end of 'while' loop.
if( dP > abs( pval.star - a2 ) ) warning( "tibberRM: did not converge for lower end-point." )
if( i >= max.iter - 1 ) warning( "tibberRM: iterations reached max.iter for lower end-point." )
L <- par.star
if( keep.iters ) {
if( is.matrix( lower.iters ) ) colnames( lower.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - (1 - alpha/2)|" )
else names( lower.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - (1 - alpha/2)|" )
out$lower.iters <- lower.iters
}
out$lower.p.value <- 1 - pval.star
out$lower.nuisance.par <- th.star
out$lower.iterations <- i
if( verbose ) cat( "\nLower bound estimate = ", L, "\n\n" )
# Do upper end-point.
if( verbose ) {
cat( "\nNow find upper bound.\n" )
cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", Ustart, "\n" )
} # end of if 'verbose' stmt.
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Ustart, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Ustart, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} # end of find starting value for parameter of upper end-point.
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
dP <- abs( pval.star - ( 1 - a2 ) )
th.star <- Ustart
if( keep.iters ) upper.iters <- c( par.star, th.star, pval.star, dP )
i <- 0
while( ( dP > tol ) && ( i < max.iter ) ) {
i <- i + 1
th.n <- th.star - step.size * ( a2 - pval.star ) / i
# if( verbose ) cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", th.n, "\n" )
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
}
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
th.star <- th.n
if( keep.iters ) upper.iters <- rbind( upper.iters, c( par.star, th.star, pval.star, abs( pval.star - a2 ) ) )
dP <- abs( pval.star - a2 )
} # end of (second) 'while' loop.
if( dP > abs( pval.star - ( 1 - a2 ) ) ) warning( "tibberRM: did not converge for upper end-point." )
if( i >= max.iter - 1 ) warning( "tibberRM: iterations reached max.iter for upper end-point." )
U <- par.star
if( keep.iters ) {
if( is.matrix( upper.iters ) ) colnames( upper.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - alpha/2|" )
else names( upper.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - alpha/2|" )
out$upper.iters <- upper.iters
}
out$upper.p.value <- 1 - pval.star
out$upper.nuisance.par <- th.star
out$upper.iterations <- i
if( verbose ) cat( "\nUpper bound estimate = ", U, "\n\n" )
obs <- do.call( statistic, c( list( data = x ), list( ... ) ) )
if( !missing( v.terms ) ) {
obs.v <- obs[ v.terms ]
obs <- obs[ -v.terms ]
out$type <- "STIB"
# warning( "tibberRM: v.terms specified, but STIB not yet supported with Robbins-Monro algorithm." )
} else out$type <- "TIB"
res <- c( L, obs, U )
names( res ) <- c( "Lower", "Estimate", "Upper" )
out$result <- res
class( out ) <- "tibRMed"
tot.tiid <- Sys.time() - begin.tiid
out$total.time <- tot.tiid
return( out )
} # end of 'tibberRM' function.
print.tibbed <- function( x, ... ) {
cat( "\nCall:\n" )
print( x$call )
cat( "\nTest-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (interpolated)\n" )
print( x$TIB )
if( !is.null( x$STIB ) ) {
cat( "Studentized version:\n" )
print( x$STIB )
}
# cat( "\nEstimated achieved p-values:\n" )
# print( x$estimated.pvalue )
invisible()
} # end of 'print.tibbed' function.
plot.tibbed <- function( x, ..., type = c("pvalue", "estimates") ) {
type <- match.arg( type )
theta1 <- x$results[ 1, ]
if( type == "pvalue" ) {
y <- x$results[ 3, ]
yl <- "Bootstrap estimated p-value"
} else {
y <- x$results[ 2, ]
yl <- "Median Parameter Estimate from Bootstrap Resamples"
}
if( type == "pvalue" && dim( x$results )[ 1 ] == 4 ) {
y2 <- x$results[ 4, ]
leg <- TRUE
} else leg <- FALSE
a <- list( ... )
if( is.null( a$xlab ) && is.null( a$ylab ) ) plot( theta1, y, xlab = "Est. Parameter", ylab = yl, ... )
else if( is.null( a$xlab ) ) plot( theta1, y, ylab = yl, ... )
else if( is.null( a$ylab ) ) plot( theta1, y, xlab = "Est. Parameter", ... )
else plot( theta1, y, ... )
if( type == "pvalue" ) abline( v = x$TIB.interpolated, h = c( x$alpha / 2, 1 - x$alpha / 2 ), col = "blue" )
if( !is.null( a$pch ) ) a.pch <- a$pch
else a.pch <- "o"
if( !is.null( a$col ) ) a.col <- a$col
else a.col <- 1
if( !is.null( a$cex ) ) a.cex <- a$cex
else a.cex <- 1
if( leg && type == "pvalue" ) {
if( a.col != "gray" ) b.col <- "gray"
else b.col <- "black"
points( theta1, y2, pch = a.pch, col = b.col, cex = a.cex )
abline( v = x$STIB.interpolated, h = c( x$alpha / 2, 1 - x$alpha / 2 ),
col = "blue", lty = 2 )
legend( "topright", legend = c( "TIB p-values (CI in vertical blue solid line)",
"STIB p-values (CI in vertical blue dashed line)" ), col = c( a.col, b.col ),
cex = a.cex, pch = a.pch, bty = "n" )
} # end of if add STIB p-values and legend to plot stmts.
} # end of 'plot.tibbed' function.
print.tibRMed <- function( x, ... ) {
cat( "\nCall:\n" )
print( x$call )
cat( "\nAchieved estimated p-value for lower = ", x$lower.p.value, "\n" )
cat( "Associated nusiance parameter = ", x$lower.nuisance.par, "\n\n" )
cat( "Achieved estimated p-value for upper = ", x$upper.p.value, "\n" )
cat( "Associated nusiance parameter = ", x$upper.nuisance.par, "\n\n" )
if( x$type == "TIB" ) cat( "Test-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (Robbins-Monro Algorithm)\n" )
else if( x$type == "STIB" ) cat( "(Studentized) Test-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (Robbins-Monro Algorithm)\n" )
print( x$result )
invisible()
} # end of 'print.tibRMed' function.
plot.tibRMed <- function( x, ..., xlab = "Estimate", ylab = "p-value" ) {
if( is.null( x$lower.iters ) ) stop( "plot: Must use keep.iters = TRUE to make plot." )
lx <- x$lower.iters[, 1 ]
ly <- x$lower.iters[, 3 ]
ux <- x$upper.iters[, 1 ]
uy <- x$upper.iters[, 3 ]
x1 <- c( lx, ux )
x2 <- c( ly, uy )
plot( x1, x2, xlab = xlab, ylab = ylab, ..., type = "n" )
points( lx, ly, pch = "l", col = "darkblue" )
points( ux, uy, pch = "u", col = "lightblue" )
abline( v = x$result, h = c( x$alpha / 2, 1 - x$alpha / 2 ), col = "blue" )
invisible()
} # end of 'plot.tibRMed' function.
Try the distillery package in your browser
Any scripts or data that you put into this service are public.
distillery documentation built on May 19, 2021, 9:08 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 plot.tibRMed print.tibRMed plot.tibbed print.tibbed tibberRM tibber
Documented in plot.tibbed plot.tibRMed print.tibbed print.tibRMed tibber tibberRM
tibber <- function( x, statistic, B, rmodel, test.pars, rsize, block.length = 1, v.terms,
shuffle = NULL, replace = TRUE, alpha = 0.05, verbose = FALSE, ... ) {
begin.tiid <- Sys.time()
theCall <- match.call()
out <- list()
#
# 'rmodel' is a random generator function that minimally takes arguments:
# 'data', 'par', 'n', and '...', where 'par' is a single parameter value
# 'n' is the sample size to draw.
#
if( missing( x ) ) stop( "tibber: Must supply x argument." )
if( missing( statistic ) ) stop( "tibber: Must supply statistic argument." )
if( missing( B ) ) stop( "tibber: Must supply B argument." )
if( missing( rmodel ) ) stop( "tibber: Must supply rmodel argument." )
if( missing( test.pars ) ) stop( "tibber: test.pars argument not supplied." )
nT <- length( test.pars )
if( missing( rsize ) ) {
if( is.null( dim( x ) ) ) n <- length( x )
else n <- dim( x )[ 1 ]
rsize <- n
} else n <- rsize
obs <- do.call( statistic, c( list( data = x ), list( ... ) ) )
if( !missing( v.terms ) ) {
if( length( obs ) > 2 ) stop( "tibber: statistic must return only one value, or one value and its variance." )
obs.v <- obs[ v.terms ]
obs <- obs[ -v.terms ]
} else if( length( obs ) > 1 ) stop( "tibber: statistic must return only one value." )
# end of 'v.terms' missing or not stmts.
tfun <- function( tp, x, st, B, rmodel, N, rsize, bl, v.terms,
shuffle, replace, o, osd, ... ) {
xstar <- do.call( rmodel, c( list( data = x, par = tp, n = N ), list( ... ) ) )
if( verbose ) cat( "Nuisance parameter = ", tp, "\n" )
if( missing( rsize ) && missing( v.terms ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, block.length = bl,
shuffle = shuffle, replace = replace, ... )
} else if( missing( rsize ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, block.length = bl,
v.terms = v.terms, shuffle = shuffle, replace = replace, ... )
} else if( missing( v.terms ) ) {
bootobj <- booter( x = xstar, statistic = st, B = B, rsize = rsize, block.length = bl,
shuffle = shuffle, replace = replace, ... )
} else {
bootobj <- booter( x = xstar, statistic = st, B = B, rsize = rsize, block.length = bl,
v.terms = v.terms, shuffle = shuffle, replace = replace, ... )
}
# Qk <- do.call( st, c( list( data = xstar[ c( bootobj$indices ) ], list( ... ) ) ) )
Qk <- bootobj$original.est
if( !missing( v.terms ) ) Qk.v <- bootobj$orig.v
# TIB is not defined for more than one parameter at a time, but for now allowing
# for the possibility.
if( is.null( dim( bootobj$results ) ) ) {
B2 <- sum( !is.na( bootobj$results ), na.rm = TRUE )
Qmk <- quantile( bootobj$results, probs = 0.5 )
Pk <- ( sum( bootobj$results < o, na.rm = TRUE ) + sum( bootobj$results == o, na.rm = TRUE ) / 2 ) / B2
res <- c( Qk, Qmk, Pk )
if( !missing( v.terms ) ) {
Qmk.sd <- sqrt( bootobj$v )
Tstar <- ( bootobj$results - Qk ) / Qmk.sd
ot <- ( o - Qk ) / osd
Pk.stud <- ( sum( Tstar < ot, na.rm = TRUE ) + sum( Tstar == ot, na.rm = TRUE ) / 2 ) / B2
# Qk.stud <- quantile( Tstar, probs = 0.5 )
res <- c( res, Pk.stud )
} # end of if do studentized intervals too stmt.
} else {
B2 <- colSums( !is.na( bootobj$results ), na.rm = TRUE )
Qmk <- apply( bootobj$results, 2, quantile, probs = 0.5 )
bigO <- matrix( o, nrow = nrow( bootobj$results ), ncol = B )
o1 <- bootobj$results < bigO
o2 <- bootobj$results == bigO
Pk <- rowSums( o1 + o2 / 2, na.rm = TRUE ) / B2
res <- c( Qk, Qmk, Pk )
if( !missing( v.terms ) ) {
Qmk.sd <- matrix( sqrt( bootobj$v ), nrow = nrow( bootobj$results ), ncol = B )
Qkmat <- matrix( Qk, nrow = nrow( bootobj$results ), ncol = B )
Tstar <- ( bootobj$results - Qkmat ) / Qmk.sd
ot <- matrix( ( o - Qk ) / osd, nrow = nrow( bootobj$results ), ncol = B )
o1 <- Tstar < ot
o2 <- Tstar == ot
Pk.stud <- ( rowSums( o1 + o2 / 2, na.rm = TRUE ) ) / B2
# Qk.stud <- apply( Tstar, 1, quantile, probs = 0.5 )
res <- c( res, Pk.stud )
} # end of if do studentized intervals too stmt.
}
return( res )
} # end of internal 'tfun' function.
if( missing( rsize ) && missing( v.terms ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B,
rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, ... )
} else if( missing( rsize ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B,
v.terms = v.terms, rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, osd = sqrt( obs.v ), ... )
} else if( missing( v.terms ) ) {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B, rsize = rsize,
rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, ... )
} else {
res <- apply( cbind( test.pars ), 1, tfun, x = x, st = statistic, B = B, rsize = rsize,
v.terms = v.terms, rmodel = rmodel, N = n, bl = block.length, shuffle = shuffle,
replace = replace, o = obs, osd = sqrt( obs.v ), ... )
}
if( missing( v.terms ) ) rownames( res ) <- c( "Est. Parameter", "Bootstrap Median Est. Parameter", "Est. P-value" )
else rownames( res ) <- c( "Est. Parameter", "Bootstrap Median Est. Parameter", "Est. P-value", "Est. P-value (Stud.)" )
colnames( res ) <- test.pars
out$results <- res
if( nT > 1 ) { # Do interpolation method.
# 'res' is a 3 (or 4 if v.terms present) by k matrix.
P <- res[ 3, ]
Q <- res[ 1, ]
a2 <- alpha / 2
q1 <- min( which( P < 1 - a2 ), na.rm = TRUE )
q2 <- max( which( P >= 1 - a2 ), na.rm = TRUE )
P1 <- P[ q1 ]
P2 <- P[ q2 ]
Q1 <- Q[ q1 ]
Q2 <- Q[ q2 ]
L <- ( Q1 - Q2 ) * ( P2 - ( 1 - a2 ) ) / ( P2 - P1 ) + Q2
out$Plow <- c( P1, P2 )
# Plow.est <- P1 + ( P2 - P1 ) * ( L - Q1 ) / ( Q2 - Q1 )
q1 <- min( which( P <= a2 ), na.rm = TRUE )
q2 <- max( which( P > a2 ), na.rm = TRUE )
P1 <- P[ q1 ]
P2 <- P[ q2 ]
Q1 <- Q[ q1 ]
Q2 <- Q[ q2 ]
U <- ( Q1 - Q2 ) * ( P2 - a2 ) / ( P2 - P1 ) + Q2
out$Pup <- c( P1, P2 )
# Note: if it is possible to have multiple statistics, then the following will need modification.
tib <- c( L, obs, U )
names( tib ) <- c( "Lower", "Estimate", "Upper" )
# Pup.est <- P1 + ( P2 - P1 ) * ( U - Q1 ) / ( Q2 - Q1 )
# pval.estimated <- c( 1 - Plow.est, 1 - Pup.est )
# names( pval.estimated ) <- c( "lower", "upper" )
out$TIB.interpolated <- tib
if( !missing( v.terms ) ) {
Pstud <- res[ 4, ]
# Qstud <- res[ 5, ]
q1stud <- min( which( Pstud < 1 - a2 ), na.rm = TRUE )
q2stud <- max( which( Pstud >= 1 - a2 ), na.rm = TRUE )
P1stud <- Pstud[ q1stud ]
P2stud <- Pstud[ q2stud ]
Q1 <- Q[ q1stud ]
Q2 <- Q[ q2stud ]
Lstud <- ( Q1 - Q2 ) * ( P2stud - ( 1 - a2 ) ) / ( P2stud - P1stud ) + Q2
out$PstudLow <- c( P1stud, P2stud )
# PstudLow <- P1stud + ( P2stud - P1stud ) * ( Lstud - Q1 ) / ( Q2 - Q1 )
q1stud <- min( which( Pstud <= a2 ), na.rm = TRUE )
q2stud <- max( which( Pstud > a2 ), na.rm = TRUE )
P1stud <- Pstud[ q1stud ]
P2stud <- Pstud[ q2stud ]
Q1 <- Q[ q1stud ]
Q2 <- Q[ q2stud ]
Ustud <- ( Q1 - Q2 ) * ( P2stud - a2 ) / ( P2stud - P1stud ) + Q2
stib <- c( Lstud, obs, Ustud )
names( stib ) <- c( "Lower", "Estimate", "Upper" )
out$STIB.interpolated <- stib
out$PstudUp <- c( P1stud, P2stud )
# PstudUp <- P1stud + ( P2stud - P1stud ) * ( Ustud - Q1 ) / (Q2 - Q1 )
# pval.estimated <- rbind( pval.estimated, c( 1 - PstudLow, 1 - PstudUp ) )
# rownames( pval.estimated ) <- c( "regular", "studentized" )
} # end of if do STIB interval stmt.
# out$estimated.pvalue <- pval.estimated
out$call <- theCall
out$data <- x
out$statistic <- statistic
out$B <- B
out$rmodel <- rmodel
out$n <- n
out$test.pars <- test.pars
out$rsize <- rsize
out$block.length <- block.length
if( !missing( v.terms ) ) out$v.terms <- v.terms
out$shuffle <- shuffle
out$replace <- replace
out$alpha <- alpha
tot.tiid <- Sys.time() - begin.tiid
out$total.time <- tot.tiid
if( verbose ) print( tot.tiid )
class( out ) <- "tibbed"
return( out )
} else return( res )
} # end of 'tibber' function.
tibberRM <- function( x, statistic, B, rmodel, startval, rsize, block.length = 1, v.terms,
shuffle = NULL, replace = TRUE, alpha = 0.05, step.size, tol = 1e-4, max.iter = 1000,
keep.iters = TRUE, verbose = FALSE, ... ) {
if( missing( startval ) ) stop( "tibberRM: must specify startval." )
begin.tiid <- Sys.time()
theCall <- match.call()
out <- list()
out$call <- theCall
out$x <- x
out$statistic <- statistic
out$B <- B
out$rmodel <- rmodel
if( !missing( v.terms ) ) out$v.terms <- v.terms
out$shuffle <- shuffle
out$alpha <- alpha
out$step.size <- step.size
out$tolerance <- tol
out$extra.args <- list( ... )
out$block.length <- block.length
out$replace <- replace
if( missing( rsize ) ) {
if( is.null( dim( x ) ) ) n <- length( x )
else n <- dim( x )[ 1 ]
rsize <- n
} else n <- rsize
out$rsize <- rsize
# Specify starting values for the nuisance parameter.
if( length( startval ) == 1 ) Lstart <- Ustart <- startval
else if( length( startval ) == 2 ) {
Lstart <- startval[ 1 ]
Ustart <- startval[ 2 ]
} else stop( "tibberRM: invalid startval argument." )
a2 <- alpha / 2
if( verbose ) {
cat( "\nBeginning to find lower bound.\n" )
cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", Lstart, "\n" )
} # end of if 'verbose' stmt.
# Do lower end-point.
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Lstart, rsize = rsize, block.length = block.length,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Lstart, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} # end of find starting value for parameter of lower end-point.
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
dP <- abs( pval.star - a2 )
th.star <- Lstart
if( keep.iters ) lower.iters <- c( par.star, th.star, pval.star, dP )
i <- 0
if( verbose ) cat( "Initial estimate (p-value) = ", par.star, "(", pval.star, ")\n" )
while( ( dP > tol ) && ( i < max.iter ) ) {
i <- i + 1
th.n <- th.star - step.size * ( ( 1 - a2 ) - pval.star ) / i
# if( verbose ) cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", th.n, "\n" )
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
}
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
th.star <- th.n
if( keep.iters ) lower.iters <- rbind( lower.iters, c( par.star, th.star, pval.star, abs( pval.star - ( 1 - a2 ) ) ) )
dP <- abs( pval.star - ( 1 - a2 ) )
} # end of 'while' loop.
if( dP > abs( pval.star - a2 ) ) warning( "tibberRM: did not converge for lower end-point." )
if( i >= max.iter - 1 ) warning( "tibberRM: iterations reached max.iter for lower end-point." )
L <- par.star
if( keep.iters ) {
if( is.matrix( lower.iters ) ) colnames( lower.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - (1 - alpha/2)|" )
else names( lower.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - (1 - alpha/2)|" )
out$lower.iters <- lower.iters
}
out$lower.p.value <- 1 - pval.star
out$lower.nuisance.par <- th.star
out$lower.iterations <- i
if( verbose ) cat( "\nLower bound estimate = ", L, "\n\n" )
# Do upper end-point.
if( verbose ) {
cat( "\nNow find upper bound.\n" )
cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", Ustart, "\n" )
} # end of if 'verbose' stmt.
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Ustart, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = Ustart, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
} # end of find starting value for parameter of upper end-point.
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
dP <- abs( pval.star - ( 1 - a2 ) )
th.star <- Ustart
if( keep.iters ) upper.iters <- c( par.star, th.star, pval.star, dP )
i <- 0
while( ( dP > tol ) && ( i < max.iter ) ) {
i <- i + 1
th.n <- th.star - step.size * ( a2 - pval.star ) / i
# if( verbose ) cat( "Obtaining bootstrap estimated p-value for nuisance parameter = ", th.n, "\n" )
if( missing( v.terms ) ) {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, shuffle = shuffle,
replace = replace, alpha = alpha, verbose = verbose, ... )
} else {
theta.star <- tibber( x = x, statistic = statistic, B = B, rmodel = rmodel,
test.pars = th.n, rsize = rsize, block.length = block.length, v.terms = v.terms,
shuffle = shuffle, replace = replace, alpha = alpha,
verbose = verbose, ... )
}
par.star <- theta.star[ 1 ]
if( missing( v.terms ) ) pval.star <- theta.star[ 3 ]
else pval.star <- theta.star[ 4 ]
th.star <- th.n
if( keep.iters ) upper.iters <- rbind( upper.iters, c( par.star, th.star, pval.star, abs( pval.star - a2 ) ) )
dP <- abs( pval.star - a2 )
} # end of (second) 'while' loop.
if( dP > abs( pval.star - ( 1 - a2 ) ) ) warning( "tibberRM: did not converge for upper end-point." )
if( i >= max.iter - 1 ) warning( "tibberRM: iterations reached max.iter for upper end-point." )
U <- par.star
if( keep.iters ) {
if( is.matrix( upper.iters ) ) colnames( upper.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - alpha/2|" )
else names( upper.iters ) <- c( "parameter", "nuisance parameter", "p-value", "|p-value - alpha/2|" )
out$upper.iters <- upper.iters
}
out$upper.p.value <- 1 - pval.star
out$upper.nuisance.par <- th.star
out$upper.iterations <- i
if( verbose ) cat( "\nUpper bound estimate = ", U, "\n\n" )
obs <- do.call( statistic, c( list( data = x ), list( ... ) ) )
if( !missing( v.terms ) ) {
obs.v <- obs[ v.terms ]
obs <- obs[ -v.terms ]
out$type <- "STIB"
# warning( "tibberRM: v.terms specified, but STIB not yet supported with Robbins-Monro algorithm." )
} else out$type <- "TIB"
res <- c( L, obs, U )
names( res ) <- c( "Lower", "Estimate", "Upper" )
out$result <- res
class( out ) <- "tibRMed"
tot.tiid <- Sys.time() - begin.tiid
out$total.time <- tot.tiid
return( out )
} # end of 'tibberRM' function.
print.tibbed <- function( x, ... ) {
cat( "\nCall:\n" )
print( x$call )
cat( "\nTest-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (interpolated)\n" )
print( x$TIB )
if( !is.null( x$STIB ) ) {
cat( "Studentized version:\n" )
print( x$STIB )
}
# cat( "\nEstimated achieved p-values:\n" )
# print( x$estimated.pvalue )
invisible()
} # end of 'print.tibbed' function.
plot.tibbed <- function( x, ..., type = c("pvalue", "estimates") ) {
type <- match.arg( type )
theta1 <- x$results[ 1, ]
if( type == "pvalue" ) {
y <- x$results[ 3, ]
yl <- "Bootstrap estimated p-value"
} else {
y <- x$results[ 2, ]
yl <- "Median Parameter Estimate from Bootstrap Resamples"
}
if( type == "pvalue" && dim( x$results )[ 1 ] == 4 ) {
y2 <- x$results[ 4, ]
leg <- TRUE
} else leg <- FALSE
a <- list( ... )
if( is.null( a$xlab ) && is.null( a$ylab ) ) plot( theta1, y, xlab = "Est. Parameter", ylab = yl, ... )
else if( is.null( a$xlab ) ) plot( theta1, y, ylab = yl, ... )
else if( is.null( a$ylab ) ) plot( theta1, y, xlab = "Est. Parameter", ... )
else plot( theta1, y, ... )
if( type == "pvalue" ) abline( v = x$TIB.interpolated, h = c( x$alpha / 2, 1 - x$alpha / 2 ), col = "blue" )
if( !is.null( a$pch ) ) a.pch <- a$pch
else a.pch <- "o"
if( !is.null( a$col ) ) a.col <- a$col
else a.col <- 1
if( !is.null( a$cex ) ) a.cex <- a$cex
else a.cex <- 1
if( leg && type == "pvalue" ) {
if( a.col != "gray" ) b.col <- "gray"
else b.col <- "black"
points( theta1, y2, pch = a.pch, col = b.col, cex = a.cex )
abline( v = x$STIB.interpolated, h = c( x$alpha / 2, 1 - x$alpha / 2 ),
col = "blue", lty = 2 )
legend( "topright", legend = c( "TIB p-values (CI in vertical blue solid line)",
"STIB p-values (CI in vertical blue dashed line)" ), col = c( a.col, b.col ),
cex = a.cex, pch = a.pch, bty = "n" )
} # end of if add STIB p-values and legend to plot stmts.
} # end of 'plot.tibbed' function.
print.tibRMed <- function( x, ... ) {
cat( "\nCall:\n" )
print( x$call )
cat( "\nAchieved estimated p-value for lower = ", x$lower.p.value, "\n" )
cat( "Associated nusiance parameter = ", x$lower.nuisance.par, "\n\n" )
cat( "Achieved estimated p-value for upper = ", x$upper.p.value, "\n" )
cat( "Associated nusiance parameter = ", x$upper.nuisance.par, "\n\n" )
if( x$type == "TIB" ) cat( "Test-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (Robbins-Monro Algorithm)\n" )
else if( x$type == "STIB" ) cat( "(Studentized) Test-inversion Bootstrap ", ( 1 - x$alpha ) * 100, "% CI (Robbins-Monro Algorithm)\n" )
print( x$result )
invisible()
} # end of 'print.tibRMed' function.
plot.tibRMed <- function( x, ..., xlab = "Estimate", ylab = "p-value" ) {
if( is.null( x$lower.iters ) ) stop( "plot: Must use keep.iters = TRUE to make plot." )
lx <- x$lower.iters[, 1 ]
ly <- x$lower.iters[, 3 ]
ux <- x$upper.iters[, 1 ]
uy <- x$upper.iters[, 3 ]
x1 <- c( lx, ux )
x2 <- c( ly, uy )
plot( x1, x2, xlab = xlab, ylab = ylab, ..., type = "n" )
points( lx, ly, pch = "l", col = "darkblue" )
points( ux, uy, pch = "u", col = "lightblue" )
abline( v = x$result, h = c( x$alpha / 2, 1 - x$alpha / 2 ), col = "blue" )
invisible()
} # end of 'plot.tibRMed' function.
Try the distillery 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.