Skip to content

Extract the procedure a nested run used

Source: R/extract-procedure.R
extract_procedure.Rd

extract_procedure() returns what a nested run recorded about its procedure: the tuner and its arguments, the selection rule, the control and the workflow. nested_final_fit() runs this procedure again.

Usage

extract_procedure(x, ...)

Arguments

x

A nested_results from nested_tune_grid() or a related function, or a nested_final_fit.

...

Not used, and must be empty.

Value

A named list with these elements, in this order:

  • tuner and the tuner's own arguments

  • param_info, event_level and eval_time

  • select and first_metric

  • control and workflow

Details

tuner names the tune or finetune function that ran, with its own arguments beside it:

  • "tune_grid", with grid

  • "tune_race_anova" or "tune_race_win_loss", also with grid

  • "tune_bayes", with iter, initial and objective

  • "tune_sim_anneal", with iter and initial

  • "fit_resamples", with none

select is the selection_rule(), and first_metric is the first metric of the set, as tune resolves it. A run given no metric set names the first metric of tune's default set. param_info, event_level and eval_time are as given. control is the control that took effect, with the settings this package forces applied. On a Bayesian run, its seed is left out. A "fit_resamples" record has no param_info, select or first_metric.

workflow describes the model, the preprocessor, and any case-weights column and tailor postprocessor as text. nested_final_fit() uses it to check that it gets the same workflow. A model argument is stored as written, so a variable's name is stored and not its value. The workflow object and the data are not stored.

See also

nested_final_fit(), selection_rule(), extract_tune_results(), nested_tune_grid() and the other loop functions for what the forced control settings are

Examples

data(mtcars)

rec <- recipes::recipe(mpg ~ ., data = mtcars) |>
  recipes::step_pca(recipes::all_predictors(), num_comp = tune::tune())
wf <- workflows::workflow(rec, parsnip::linear_reg())

set.seed(1)
folds <- nested_resamples(
  mtcars,
  outside = rsample::vfold_cv(v = 2),
  inside = rsample::vfold_cv(v = 2)
)
set.seed(2)
res <- nested_tune_grid(wf, folds, grid = data.frame(num_comp = 1:2))
set.seed(3)
final <- nested_final_fit(wf, res)
procedure <- extract_procedure(res)
procedure$tuner
#> [1] "tune_grid"
procedure$control$allow_par
#> [1] FALSE

identical(extract_procedure(final)$tuner, procedure$tuner)
#> [1] TRUE