Skip to contents

Fit the Configuration Ranked `.rank` in a Workflow Set

Source: R/tuning_results.R
workflow_set_rank_fit.Rd

Ranks the configurations of tuned workflow set results by `rank_metric`, picks the one at position `.rank`, and fits it on the full training set with [tune::fit_best()].

Usage

workflow_set_rank_fit(
  grid_results,
  rank_metric,
  .rank = 1,
  group_wflow = TRUE,
  select_best = FALSE,
  ...
)

Arguments

grid_results

Tuned workflow set results, the output of [workflowsets::workflow_map()] run with `control = tune::control_grid(save_workflow = TRUE)`, which [tune::fit_best()] needs.

rank_metric

Name of the metric to rank by.

.rank

Position to fit; 1 (default) is the best.

group_wflow

If `TRUE` (default), rank workflows, each by its best configuration, so `.rank = 2` is the second-best *workflow*. If `FALSE`, rank configurations directly, so `.rank = 2` can be a second configuration of the best workflow.

select_best

Passed to [workflowsets::rank_results()]: keep only each workflow's best configuration before ranking.

...

Passed to [tune::fit_best()].

Value

A list with `model`, the fitted workflow, and `tuned_parameters`, a one-row tibble of the workflow's `wflow_id`, `preproc`, `model`, tuning parameters, and its [workflowsets::rank_results()] row (`.config`, `.metric`, `mean`, `std_err`, `n`, `rank`, and `overall_rank` when `group_wflow = TRUE`).

See also

[workflow_set_params()]

Examples

if (FALSE) { # \dontrun{
results <- workflowsets::workflow_map(
  wset, "tune_grid", resamples = folds, grid = 5,
  control = tune::control_grid(save_workflow = TRUE)
)
best <- workflow_set_rank_fit(results, rank_metric = "rmse")
best$tuned_parameters
predict(best$model, new_data)
} # }