Clean Up Mismatched or Superseded Checkpoint Files (E5)

Inspects `.rds` checkpoint files in `dir` and removes files from older runs, mismatched fingerprints, or superseded `seed_scheme` versions while keeping the active run's checkpoints.

Usage

clean_checkpoints(
  dir,
  keep = c("current", "matching", "none"),
  run_fingerprint = NULL,
  seed_scheme = NULL,
  recursive = FALSE,
  dry_run = FALSE
)

Arguments

dir

Directory containing checkpoint `.rds` files.

keep

Retention policy:

`"current"`

(Default) Keeps checkpoint files belonging to the most recent `checkpoint_run_fingerprint` in each checkpoint family (or the explicitly supplied `run_fingerprint`) and removes older, mismatched, or corrupted checkpoint files.

`"matching"`

Keeps only checkpoint files matching the supplied `run_fingerprint` and/or `seed_scheme`.

`"none"`

Removes all checkpoint files in `dir`.

run_fingerprint

Optional expected `checkpoint_run_fingerprint` hash. When `NULL` under `keep = "current"`, the fingerprint of the newest valid checkpoint file per family (`tune` / `score_outer`) is used.

seed_scheme

Optional expected seed scheme string (e.g. `"v0.5.6"`).

recursive

Logical (default `FALSE`). Search subdirectories of `dir`.

dry_run

Logical (default `FALSE`). When `TRUE`, reports which files would be kept or removed without deleting anything from disk.

Value

A tibble of class `cbe_clean_checkpoints` with columns `file`, `path`, `family`, `run_fingerprint`, `seed_scheme`, `mtime`, `action` (`"kept"` or `"removed"`), and `reason`, plus attributes `n_kept` and `n_removed`.

See also

[collect_checkpoints()], [tune_alpha_by_split()], [score_outer()]