AUC#

class scio.eval.AUC(*, max_fpr=1.0, min_fpr=0.0, kind='pessimistic')[source]#

AUC for ROC, potentially partial — in which case normalized.

With the default arguments, one has

\[AUC = \mathbb{P}(\text{score}_{\text{OoD}}<\text{score}_{\text{InD}}),\]

when sampling from the reference population.

Parameters:

max_fpr (float) – Maximum False Positive Rate for partial AUC. Defaults to 1.0.
min_fpr (float) – Minimum False Positive Rate for partial AUC. Defaults to 0.0.
kind (InterpolationKindLike) – See InterpolationKind. Defaults to "pessimistic".