Score Candidate Models on Outer Resampling Split(s)
Source:R/compare_censoring_weights.R
score_outer.RdA tidy wrapper around [glmnet_IBS()] using [furrr::future_map()] (or sequential [purrr::map()]) to evaluate a grid of candidate models across one or more outer resampling splits.
Usage
score_outer(
object,
grid,
recipe,
...,
censoring_weights = c("ipcw", "none"),
coefficients = c("all", "nonzero"),
seed = NULL,
checkpoint_dir = NULL,
checkpoint_mismatch = c("recompute", "overwrite", "error"),
shard = NULL,
chunk_size = NULL,
on_complete = getOption("templecbe.on_complete", NULL),
progress = FALSE
)Arguments
- object
An `rsplit` (e.g. from [rsample::initial_split()]), an `rset` / [rsample::nested_cv()] object containing outer splits, or a list of `rsplit` objects.
- grid
A data frame or tibble defining candidate models to evaluate. Must contain an `alpha` column. If a `formula` or `formulas` column is present, each row's formula is passed to [glmnet_IBS()]. An optional `model` or `model_id` column will be preserved as the model identifier; if absent, one will be created.
- recipe
A recipe object specifying preprocessing.
- ...
Additional arguments passed to [glmnet_IBS()] (such as `time_data`, `internal_folds`, `feature_names`, `id_col`, `cox.ties`, `covariates`, `rule`, `metric`, `tune_engine`).
- censoring_weights
Weighting scheme: `"ipcw"` (default) or `"none"`.
- coefficients
Option controlling which coefficients to store in the `coefficients` list column: `"all"` (default, includes predictors shrunk to 0) or `"nonzero"` (filters to non-zero estimates).
- seed
Optional integer seed for reproducible model fitting across future plans.
- checkpoint_dir
Optional directory path to save per-task checkpoint files. Finished tasks are loaded from disk on rerun when their fingerprint matches, skipping recomputation.
- checkpoint_mismatch
Action when a checkpoint file in `checkpoint_dir` has a mismatched fingerprint: `"recompute"` (default; warns and recomputes), `"overwrite"` (recomputes and overwrites mismatched checkpoint files in place without warning; can also be enabled via `options(templecbe.checkpoint_mismatch = "overwrite")` or `options(templecbe.checkpoint_overwrite = TRUE)`), or `"error"` (stops immediately).
- shard
Optional integer vector `c(i, n)` specifying that this machine should evaluate shard `i` of `n` into `checkpoint_dir`. Use [collect_checkpoints()] to merge all shards once finished.
- chunk_size
Optional integer number of tasks per worker chunk passed to [furrr::furrr_options()] (default `NULL`, which lets furrr divide tasks evenly into one chunk per worker).
- on_complete
Optional callback `function(result)` invoked when the run finishes or errors (defaults to `getOption("templecbe.on_complete", NULL)`).
- progress
Show progress bar (default `FALSE`).