Extract the tuning run a final fit was selected from

Returns the tune::tune_grid() result that nested_final_fit() chose its parameters from — the record of what selection saw when the procedure was re-run on the complete dataset.

Usage

extract_tune_results(x, ...)

Arguments

x

A nested_final_fit object from nested_final_fit().

...

Not used.

Value

The stored tune_results object, unchanged. It is tune's own object, so tune's generics apply to it directly.

What its numbers are, and are not

The returned object answers collect_metrics(), and will hand its metrics over without qualifying them. Every one of them is a selection-time quantity: it was computed on the resamples that chose the candidate it describes, which makes it optimistically biased as a claim about the model this final fit produced. Nothing in that object is this model's performance.

Report the nested estimate instead — collect_metrics() on the nested_tune_grid() result. That number is measured on data no part of the tune-and-fit procedure ever saw, which is what makes it an honest description of the procedure that produced your model.

The run is kept because it is the record of what selection saw, not because it describes the model.

See also

extract_scored_candidates(), nested_final_fit(), nested_tune_grid()

Examples

data(mtcars)

rec <- recipes::step_pca(
  recipes::recipe(mpg ~ ., data = mtcars),
  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 = 3),
  inside = rsample::vfold_cv(v = 3)
)

set.seed(3)
final <- nested_final_fit(wf, folds, grid = data.frame(num_comp = 1:3))

extract_tune_results(final)
#> # Tuning results
#> # 3-fold cross-validation 
#> # A tibble: 3 × 4
#>   splits          id    .metrics         .notes          
#>   <list>          <chr> <list>           <list>          
#> 1 <split [21/11]> Fold1 <tibble [6 × 5]> <tibble [0 × 4]>
#> 2 <split [21/11]> Fold2 <tibble [6 × 5]> <tibble [0 × 4]>
#> 3 <split [22/10]> Fold3 <tibble [6 × 5]> <tibble [0 × 4]>