Reproduce the archived benchmark grouped folds

Groups are enumerated in first-row order, classified by rounded mean outcome, shuffled within outcome (0 then 1), and assigned round-robin.

Usage

os_make_benchmark_folds(y, group_id, k = 5L, seed = 42L)

Arguments

y

Numeric binary outcome vector (0 and 1).

group_id

Nonmissing group identifiers, one per row.

k

Number of folds, at least two.

seed

Random seed.

Value

An unnamed list of sorted integer test-row indices.

Details

This reproduces the benchmark harness algorithm and changes the global RNG state through set.seed(seed). A mixed-outcome group with mean 0.5 is assigned to class 0. Supply original row order and the same RNGkind settings (the archive uses the R default). The existing os_make_grouped_stratified_folds balances row counts and retains its different partitions and mixed-group checks.