Nothing
#' CI for weighted AUC (wAUC) by replicate weights
#'
#' @description
#' A function to estimate a confidence interval (CI) for wAUC implemented by \code{wROC::wauc()} with replicate weights by \code{survey::as.svrepdesign()}; \code{bootstrap} is Canty and Davison's bootstrap,
#' \code{subbootstrap} is Rao and Wu's (n-1) bootstrap, and \code{JK1} and \code{JKn} are jackknife methods.
#'
#'
#' @param y A vector of response variable or a string of its name (requires \code{data} to be supplied).
#' @param yhat A vector of predicted probabilities of `y` or a string of its name (requires \code{data} to be supplied).
#' @param tag.event A character of the event of interest in `y`, e.g. disease, or condition. Default is \code{"1"}.
#' @param tag.nonevent A character of the non-event in `y`, e.g. non-disease, or non-condition. Default is \code{"0"}.
#' @param cluster A character string of a cluster identifier. It could be \code{NULL} if the sampling design were plugged in the \code{design} argument.
#' @param strata A character string of a strata identifier. It could be \code{NULL} if the sampling design were plugged in the \code{design} argument.
#' @param weights A character string of a sampling weights' identifier. \code{NULL} for either a unweighted case or the sampling design inserted in the \code{design} argument.
#' @param design An object of class \code{survey.design} generated by \code{survey::svydesign()}. It could be \code{NULL} if information about \code{cluster}, \code{strata}, \code{weights} and \code{data} were given.
#' @param data A data frame including vectors of \code{y}, \code{yhat}, and (or) \code{weights}. \code{cluster},\code{strata} can be included depending on the complex sampling structure.
#' \code{weights = NULL} assumes equal probability generating `1's`.
#' If \code{data=NULL}, then specific numerical vectors must be included in \code{y}, \code{yhat}, or
#' the sampling design should be plugged into the argument \code{design}.
#' @param type Type of replicate weights. Choose one of \code{"JK1"}, \code{"JKn"}, \code{"bootstrap"}, or
#' \code{"subbootstrap"}. Note: \code{JKn} for stratified, \code{JK1} for unstratified designs.
#' @param B The number of bootstrap replicates. Default is 2000.
#' @param conf.level The width of confidence interval. Default is 0.95.
#' @param fpc,fpctype Passed to \code{jk1weights, jknweights, brrweights, bootweights, subbootweights}, or \code{mrbweights}.
#' @param compress Use a compressed representation of the replicate weights matrix.
#' @param ... optional parameters to be passed to the low level function \code{survey::as.svrepdesign()}.
#' @param mse if TRUE, compute variances from sums of squares around the point estimate, rather than the mean of the replicates.
#'
#' @return confidence interval (CI) estimates
#'
#' @seealso [svyROC::wauc()] for detailed information on the implementation.
#' @seealso [survey::as.svrepdesign()] for detailed information on replicate weights methods.
#'
#'
#' @examples
#' \donttest{
#' data(nhanes2013_sbc)
#'
#' # For ad hoc adjustment for a single PSU stratum not to contribute to the variance
#' options(survey.lonely.psu='remove')
#'
#' # yhat would typically come from a fitted model (e.g. wElnet(), wXGBoost(),
#' # wRandomforest(), or any other classifier). Here we generate one directly
#' # from a simple weighted logistic regression for a self-contained example.
#' fit <- glm(HBP ~ INDFMPIR + BMXBMI + BMXARML + BMXLEG + BPXPLS,
#' data = nhanes2013_sbc,
#' family = binomial(), weights = nhanes2013_sbc$WTSAF2YR)
#' yhat <- predict(fit, type = "response")
#'
#' # Conveniently, update a dataset including yhat (a vector) estimated
#' # from any machine learning methods. HBP, SDMVPSU, SDMVSTRA, and WTSAF2YR
#' # are referenced by name (not position) to avoid any risk of selecting
#' # the wrong columns.
#' data1 <- data.frame(
#' nhanes2013_sbc[, c("HBP", "SDMVPSU", "SDMVSTRA", "WTSAF2YR")],
#' yhat
#' )
#'
#' # For estimating bootstrap 95\% CI for a model selected by dCV
#' wAUC.ci<- ci_wauc("HBP", "yhat",weights="WTSAF2YR",cluster="SDMVPSU",strata="SDMVSTRA",
#' data=data1,type="subbootstrap",B=2000)
#' #Or equivalently,
#'
#' des <- survey::svydesign(ids=~SDMVPSU, strata = ~SDMVSTRA, weights = ~WTSAF2YR,
#' nest = TRUE, data = nhanes2013_sbc)
#'
#' wAUC.ci<- ci_wauc(des$variables$HBP,data1$yhat,design = des,type = "subbootstrap",B=2000)
#'}
#' @export
# library(wROC)
# library(survey)
# for ad hoc adjustment for a single PSU stratum:
# options(survey.lonely.psu='remove')
### weights,cluster,strata are names in data (character) unless design is included.
#set.seed(234)
ci_wauc<- function(y,yhat,tag.event = "1", tag.nonevent = "0",weights=NULL,cluster=NULL,strata=NULL,design=NULL,data=NULL,
type=c("JK1","JKn", "bootstrap", "subbootstrap"),B=2000,conf.level=0.95,
fpc=NULL,fpctype=NULL,..., compress=TRUE,
mse=getOption("survey.replicates.mse")){
type <- match.arg(type)
# Step -1: Resolve y/yhat if supplied as column-name strings rather than
# vectors. This activates what was previously documented (@param y/@param
# yhat both state "or a string of its name") but not implemented -- y and
# yhat were always passed straight through to svyROC::wauc() regardless of
# type, which silently misbehaved if a bare string was passed instead of
# the vector it named.
if(inherits(y, "character") && length(y) == 1){
if(is.null(data)){
stop("'y' was supplied as a column name (a single string), but 'data' is NULL. ",
"Either supply 'data' containing a column named '", y, "', or pass 'y' directly as a vector.")
}
y <- data[, y]
}
if(inherits(yhat, "character") && length(yhat) == 1){
if(is.null(data)){
stop("'yhat' was supplied as a column name (a single string), but 'data' is NULL. ",
"Either supply 'data' containing a column named '", yhat, "', or pass 'yhat' directly as a vector.")
}
yhat <- data[, yhat]
}
if(length(table(y)) != 2){
stop("Response variable 'y' must have exactly two classes.")
}
# Step 0: Notation
if(!is.null(design)){
cluster <- as.character(design$call$id[2])
if(cluster == "1" || cluster == "0"){
cluster <- NULL
}
if(is.null(design$call$strata[2])){
strata <- NULL
} else{
strata <- as.character(design$call$strata[2])
}
if(is.null(design$call$weights) || sum(unique(design$allprob))==1){
weights <- NULL
} else{
weights <- as.character(design$call$weights[2])
}
data <- design$variables
} else{ # Define survey design
if (is.null(strata)){
if (is.null(cluster)) {
formula.cluster <- as.formula("~1")} else{
formula.cluster <- as.formula(paste0("~", cluster))
}
#generating unweighted design: assuming equal probability
if(is.null(weights)) {
design<- survey::svydesign(ids = formula.cluster,data = data)
} else{
formula.weights <- as.formula(paste0("~", weights))
design<- survey::svydesign(ids = formula.cluster, weights= formula.weights,data = data )
}
} else{
formula.strata <- as.formula(paste0("~", strata))
if(is.null(cluster)){stop("Error. No cluster supplied.")}
formula.cluster <- as.formula(paste0("~", cluster))
# BUG FIX: previously this branch unconditionally required 'weights'
# to be non-NULL whenever 'strata' was supplied, even though an
# unweighted stratified design (equal-probability sampling within
# strata) is a legitimate survey design -- the strata-less branch
# above already correctly supported weights = NULL; this branch is
# now brought into line with it.
if(is.null(weights)){
design <- survey::svydesign(ids = formula.cluster,
strata = formula.strata,
data = data, nest=TRUE)
} else {
formula.weights <- as.formula(paste0("~", weights))
design <- survey::svydesign(ids = formula.cluster,
strata = formula.strata,
weights = formula.weights,
data = data, nest=TRUE)
}
}
}
# Generate replicate weights based on the selected method
if(type %in% c("bootstrap", "subbootstrap")){
rep.des <- survey::as.svrepdesign(design = design, type = type, replicates = B)
} else {
rep.des <- survey::as.svrepdesign(design = design, type = type)
B<-dim(rep.des$repweights$weights)[2]
}
rep.weights.raw <- rep.des$repweights$weights
# BUG FIX: rep.des$repweights$weights is sized by the number of
# clusters/PSUs, NOT the number of observations in the original data
# (e.g. 12 rows for 12 PSUs, vs. 240 observations). Using it directly
# against y/yhat (length = nrow(data)) either errors outright
# (svyROC::wauc()'s own length check, as seen when weights = NULL) or,
# worse, silently "succeeds" via R's vector recycling whenever
# nrow(data) happens to be a clean multiple of the number of clusters
# -- which recycles cluster-level weights in the wrong, arbitrary
# cycling order rather than mapping each unit to the weight of the
# cluster it actually belongs to, silently corrupting every
# downstream wAUC replicate. The correct approach (matching the
# equivalent expansion in error.f(), used by wElnet()) is to expand
# the cluster-level matrix to unit-level via
# rep.des$repweights$index, which records which cluster each
# observation belongs to.
rep.weights <- apply(rep.weights.raw, 2, function(x){x[rep.des$repweights$index]})
wauc.B<- rep(0,B)
if(is.null(weights)){
for(b in 1:B){
wauc.B[b]<- svyROC::wauc(response.var=y, phat.var=yhat, weights.var = rep.weights[,b],
tag.event = tag.event, tag.nonevent = tag.nonevent)$AUCw
}
} else{
# Compute wAUC replicates by the selected method
WTs<-design$variables[,weights]
for(b in 1:B){
wauc.B[b]<- svyROC::wauc(response.var=y, phat.var=yhat, weights.var = WTs*rep.weights[,b],
tag.event = tag.event, tag.nonevent = tag.nonevent)$AUCw
}
}
CI= quantile(wauc.B,c(0+(1-conf.level)/2,.5,1-(1-conf.level)/2))
return(CI)
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.