Select Candidate Models From Tuning Results

Takes results from [tune_alpha_by_split()] or [tune_over_alpha()], identifies the best-performing `alpha` values (e.g. within the top `ntile` of integrated Brier scores), and crosses them with candidate formulas to construct an outer evaluation grid.

Usage

select_candidate_models(
  tune_results,
  formulas,
  ntile = 0.25,
  top_n = NULL,
  by = c("row", "alpha")
)

Arguments

tune_results

A data frame containing tuning results with at least `alpha` and `IBS` columns (e.g. from [tune_alpha_by_split()]).

formulas

Character vector of formulas, list of formula objects, or a single formula to evaluate.

ntile

Numeric quantile cutoff (between 0 and 1, default `0.25`) for selecting the best alphas with lowest IBS.

top_n

Optional integer number of best alphas to keep (overrides `ntile` if provided).

by

Selection granularity: `"row"` (default, selects alphas appearing in the top `ntile` of individual tune result rows, e.g. `ntile(IBS, 5) == 1`) or `"alpha"` (summarizes mean IBS across all rows per alpha before selecting).

Value

A tibble with columns `model`, `alpha`, and `formula`.

See also

[score_outer()], [compare_censoring_weights()], [tune_alpha_by_split()]