Plot cross-validated ROC curves

Draws one curve per resampling fold plus the pooled curve across all of them.

Usage

plot_roc_curve(predictions, path = NULL)

Arguments

predictions

the predictions element of a fit_patch_model() result

path

where to write a PNG; NULL returns the plot instead

Value

the plot object, or path invisibly when written to disk

Details

Per-fold curves are shown rather than only the pooled one because they carry information the summary AUC hides: a model whose folds agree closely is a different proposition from one averaging the same AUC out of wildly varying folds, and the second is not trustworthy at a single station even though both report the same number.

The curves come from held-out predictions, so they describe performance on data the model did not see. A curve drawn on training data would sit much closer to the corner and mean nothing.

Examples

if (FALSE) { # \dontrun{
result <- run_taupatch("inst/configs/mock_test.yaml")
plot_roc_curve(result$model$predictions)
} # }