Skip to contents

Repeated Blocked Cross-Validation

Source: R/borg_repeated_cv.R
borg_repeated_cv.Rd

Repeats borg_cv multiple times with different random seeds and aggregates the results. Repeated CV provides better variance estimates and more stable performance metrics than a single CV run, particularly important for spatial/temporal blocking where fold assignment depends on the random seed.

Usage

borg_repeated_cv(
  data,
  repeats = 10L,
  v = 5L,
  seeds = NULL,
  aggregate = TRUE,
  ...
)

Arguments

data

A data frame.

repeats

Integer. Number of repetitions. Default: 10.

v

Integer. Number of folds per repetition. Default: 5.

seeds

Integer vector of length repeats, or NULL (auto-generated). Controls fold assignment for each repetition.

aggregate

Logical. If TRUE (default), returns aggregated metrics. If FALSE, returns all individual fold results.

...

Additional arguments passed to borg_cv.

Value

A list with class "borg_repeated_cv" containing:

folds

List of length repeats, each element a list of v train/test fold pairs.

repeats

Number of repetitions.

v

Number of folds.

strategy

CV strategy used.

seeds

Seeds used for each repetition.

diagnosis

The BorgDiagnosis from the first run.

Has print() and autoplot() methods.

Details

Each repetition calls borg_cv() with a different seed, producing different fold assignments. The diagnosis is computed once and reused across repetitions for consistency.

For aggregation with a model, use borg_fold_performance() on each repetition's folds and combine the results.

See also

borg_cv, borg_fold_performance

Examples

set.seed(42)
d <- data.frame(
  x = runif(200), y = runif(200), z = rnorm(200)
)
rcv <- borg_repeated_cv(d, repeats = 3, v = 5,
                        coords = c("x", "y"), target = "z")
rcv
length(rcv$folds)  # 3 repetitions
length(rcv$folds[[1]])  # 5 folds each