% Generated by roxygen2: do not edit by hand

% Please edit documentation in R/stan.R

\name{kfold.hsstan}

\alias{kfold.hsstan}

\alias{kfold}

\title{K-fold cross-validation}

\usage{

\method{kfold}{hsstan}(

  x,

  folds,

  chains = 1,

  store.fits = TRUE,

  cores = getOption("mc.cores", 1),

  ...

)

}

\arguments{

\item{x}{An object of class \code{hsstan}.}

\item{folds}{Integer vector with one element per observation indicating the

cross-validation fold in which the observation should be withdrawn.}

\item{chains}{Number of Markov chains to run. By default this is set to 1,

independently of the number of chains used for \code{x}.}

\item{store.fits}{Whether the fitted models for each fold should be stored

in the returned object (\code{TRUE} by default).}

\item{cores}{Number of cores to use for parallelization (the value of

\code{options("mc.cores")} by default). The cross-validation folds will

be distributed to the available cores, and the Markov chains for each

model will be run sequentially.}

\item{...}{Further arguments passed to \code{\link[rstan:stanmodel-method-sampling]{rstan::sampling()}}.}

}

\value{

An object with classes \code{kfold} and \code{loo} that has a similar structure as the

objects returned by \code{\link[=loo]{loo()}} and \code{\link[=waic]{waic()}} and is compatible with the

\code{\link[loo]{loo_compare}} function for

comparing models. The object contains the following fields:

\item{estimates}{a matrix containing point estimates and standard errors of

the expected log pointwise predictive density ("elpd_kfold"),

the effective number of parameters ("p_kfold", always \code{NA}) and the

K-fold information criterion "kfoldic" (which is \code{-2 * elpd_kfold},

i.e., converted to the deviance scale).}

\item{pointwise}{a matrix containing the pointwise contributions of

"elpd_kfold", "p_kfold" and "kfoldic".}

\item{fits}{a matrix with two columns and number of rows equal to the number

of cross-validation folds. Column \code{fit} contains the fitted

\code{hsstan} objects for each fold, and column \code{test.idx} contains

the indices of the withdrawn observations for each fold. This is not

present if \code{store.fits=FALSE}.}

\item{data}{the dataset used in fitting the model (before withdrawing

observations). This is not present if \code{store.fits=FALSE}.}

}

\description{

Perform K-fold cross-validation using the same settings used when fitting

the model on the whole data.

}

\examples{

\donttest{

\dontshow{utils::example("hsstan", echo=FALSE)}

# continued from ?hsstan

# only 2 folds for speed of example

folds <- rep(1:2, length.out=length(df$Y))

cv.biom <- kfold(hs.biom, folds=folds, cores=2)

}

}