Skip to contents

Control cross-validation for greedy CI model tuning

Source: R/predict_ci_tree_terminal_mean.R
control_ci_tune.Rd

Creates a control object for tune_ci_tree() and tune_ci_forest(). The defaults keep tuning sequential and memory-light, while optional settings can save validation predictions, fitted fold models, extraction results, and request future-based parallel execution when the user has configured a future backend externally.

Usage

control_ci_tune(
  verbose = FALSE,
  allow_par = TRUE,
  parallel_over = c("resamples", "everything"),
  save_pred = FALSE,
  save_fit = FALSE,
  extract = NULL,
  pkgs = NULL,
  num_cores = 1L
)

Arguments

verbose

Logical; print fold-level progress.

allow_par

Logical; allow future-based parallel execution when num_cores > 1. This requires the suggested package future.apply.

parallel_over

Parallelization strategy. "resamples" and "everything" are currently accepted; both are executed over grid/resample tasks.

save_pred

Logical; save validation predictions for ci_collect_predictions().

save_fit

Logical; save fitted fold-level models.

extract

Optional function applied to each successfully fitted fold-level model. For forests, the function receives a list with forest and surrogate components.

pkgs

Optional character vector of packages to load on workers. Kept for API compatibility with parallel workflows.

num_cores

Compatibility gate for parallel execution. Worker counts are controlled by the user's active future plan.

Value

A control_ci_tune object.