rsample (version 1.2.1)

vfold_cv: V-Fold Cross-Validation

Description

V-fold cross-validation (also known as k-fold cross-validation) randomly splits the data into V groups of roughly equal size (called "folds"). A resample of the analysis data consists of V-1 of the folds while the assessment set contains the final fold. In basic V-fold cross-validation (i.e. no repeats), the number of resamples is equal to V.

Usage

vfold_cv(data, v = 10, repeats = 1, strata = NULL, breaks = 4, pool = 0.1, ...)

Value

A tibble with classes vfold_cv, rset, tbl_df, tbl, and data.frame. The results include a column for the data split objects and one or more identification variables. For a single repeat, there will be one column called id that has a character string with the fold identifier. For repeats, id is the repeat number and an additional column called id2

that contains the fold information (within repeat).

Arguments

data

A data frame.

v

The number of partitions of the data set.

repeats

The number of times to repeat the V-fold partitioning.

strata

A variable in data (single character or name) used to conduct stratified sampling. When not NULL, each resample is created within the stratification variable. Numeric strata are binned into quartiles.

breaks

A single number giving the number of bins desired to stratify a numeric stratification variable.

pool

A proportion of data used to determine if a particular group is too small and should be pooled into another group. We do not recommend decreasing this argument below its default of 0.1 because of the dangers of stratifying groups that are too small.

...

These dots are for future extensions and must be empty.

Details

With more than one repeat, the basic V-fold cross-validation is conducted each time. For example, if three repeats are used with v = 10, there are a total of 30 splits: three groups of 10 that are generated separately.

With a strata argument, the random sampling is conducted within the stratification variable. This can help ensure that the resamples have equivalent proportions as the original data set. For a categorical variable, sampling is conducted separately within each class. For a numeric stratification variable, strata is binned into quartiles, which are then used to stratify. Strata below 10% of the total are pooled together; see make_strata() for more details.

Examples

vfold_cv(mtcars, v = 10)
vfold_cv(mtcars, v = 10, repeats = 2)

library(purrr)
data(wa_churn, package = "modeldata")

set.seed(13)
folds1 <- vfold_cv(wa_churn, v = 5)
map_dbl(
  folds1$splits,
  function(x) {
    dat <- as.data.frame(x)$churn
    mean(dat == "Yes")
  }
)

set.seed(13)
folds2 <- vfold_cv(wa_churn, strata = churn, v = 5)
map_dbl(
  folds2$splits,
  function(x) {
    dat <- as.data.frame(x)$churn
    mean(dat == "Yes")
  }
)

set.seed(13)
folds3 <- vfold_cv(wa_churn, strata = tenure, breaks = 6, v = 5)
map_dbl(
  folds3$splits,
  function(x) {
    dat <- as.data.frame(x)$churn
    mean(dat == "Yes")
  }
)