create_folds: Create Folds

In splitTools: Tools for Data Splitting

Create Folds

Description

This function provides a list of row indices used for k-fold cross-validation (basic, stratified, grouped, or blocked). Repeated fold creation is supported as well. By default, in-sample indices are returned.

Usage

create_folds(
  y,
  k = 5L,
  type = c("stratified", "basic", "grouped", "blocked"),
  n_bins = 10L,
  m_rep = 1L,
  use_names = TRUE,
  invert = FALSE,
  shuffle = FALSE,
  seed = NULL
)

Arguments

y	Either the variable used for "stratification" or "grouped" splits. For other types of splits, any vector of the same length as the data intended to split.
k	Number of folds.
type	Split type. One of "stratified" (default), "basic", "grouped", "blocked".
n_bins	Approximate numbers of bins for numeric y (only for type = "stratified").
m_rep	How many times should the data be split into k folds? Default is 1, i.e., no repetitions.
use_names	Should folds be named? Default is TRUE.
invert	Set to TRUE in order to receive out-of-sample indices. Default is FALSE, i.e., in-sample indices are returned.
shuffle	Should row indices be randomly shuffled within folds? Default is FALSE.
seed	Integer random seed.

Details

By default, the function uses stratified splitting. This will balance the folds regarding the distribution of the input vector y. (Numeric input is first binned into n_bins quantile groups.) If type = "grouped", groups specified by y are kept together when splitting. This is relevant for clustered or panel data. In contrast to basic splitting, type = "blocked" does not sample indices at random, but rather keeps them in sequential groups.

Value

If invert = FALSE (the default), a list with in-sample row indices. If invert = TRUE, a list with out-of-sample indices.

See Also

partition(), create_timefolds()

Examples

y <- rep(c(letters[1:4]), each = 5)
create_folds(y)
create_folds(y, k = 2)
create_folds(y, k = 2, m_rep = 2)
create_folds(y, k = 3, type = "blocked")

splitTools documentation built on June 7, 2023, 6:25 p.m.