Flatten the optimisation trials recorded on a module into a tidy data frame containing per-trial metric summaries. Useful for producing tables or visualisations comparing trial performance. When yardstick metrics are supplied, the function also computes those metrics for each trial using the stored evaluation datasets.
Arguments
- module
A DSPrrr module optimised with
optimize_grid().- metrics
Optional yardstick metric (or metric set) to compute for each trial.
- truth
Column name (string) containing the ground-truth labels when computing yardstick metrics.
- estimate
Column name (string) containing the model predictions when computing yardstick metrics.
- ...
Additional arguments passed to yardstick metrics.
Value
A tibble with one row per trial containing columns:
trial_id- trial identifier.score- overall score recorded for the trial.mean_score,median_score,std_dev- summary statistics across the evaluation scores.n_evaluated,n_errors- counts reported by the evaluation.params- list-column with the parameters evaluated in the trial.scores- list-column with the raw per-example scores (if available).yardstick- list-column containing yardstick metric results when requested.
Examples
if (FALSE) { # \dontrun{
trial_metrics <- module_metrics(my_module)
yardstick_metrics <- module_metrics(
my_module,
metrics = yardstick::metric_set(yardstick::accuracy),
truth = target,
estimate = result
)
} # }