Collect and Merge Sharded Checkpoints From `checkpoint_dir`

Merges per-task checkpoint files written by [tune_alpha_by_split()] or [score_outer()] across multiple machines (`shard = c(i, n)`). Verifies that all checkpoint files carry the same `checkpoint_run_fingerprint` and `seed_scheme` so results from different data, recipes, feature sets, seed schemes, or settings cannot be combined by mistake.

Usage

collect_checkpoints(
  checkpoint_dir,
  type = c("auto", "tune", "score_outer"),
  allow_incomplete = FALSE
)

Arguments

checkpoint_dir

Directory containing `.rds` checkpoint files.

type

Optional checkpoint type to collect: `"auto"` (default; inferred from files in `checkpoint_dir`), `"tune"` ([tune_alpha_by_split()]), or `"score_outer"` ([score_outer()]).

allow_incomplete

Logical (default `FALSE`). When `FALSE`, stops with an error if any tasks from the full run grid are missing.

Value

A merged tibble matching the return format of [tune_alpha_by_split()] or [score_outer()].

See also

[tune_alpha_by_split()], [score_outer()], [clean_checkpoints()]