src.main.viyog_metrics

src.main.viyog_metrics(id_scores, ood_scores, recall_level=0.95)[source]

Compute a collection of OOD detection metrics from id and ood scores.

Parameters:

id_scores (array-like) – Scores for in-distribution examples (lower means ID in the code’s labeling).
ood_scores (array-like) – Scores for OOD examples.
recall_level (float, optional) – TPR level for which to compute FPR (default 0.95).

Returns:

Dictionary with keys “AUROC”, “AUPR_IN”, “AUPR_OUT”, “FPR95”, “DetectionError”, “AUTC” and “AUTC_components”.

Return type:

dict