SCODRiskAtxCov#

class torch_uncertainty.metrics.classification.SCODRiskAtxCov(cov_threshold, ood_cost=0.5, **kwargs)[source]#

Calculate the SCOD risk at a specified coverage.

This metric applies RiskAtxCov to the joint SCOD loss defined by SCODAURC.

For a target coverage \(\gamma\in[0,1]\), let

\[k_\gamma = \left\lceil \gamma N \right\rceil.\]

Samples are ordered from lowest to highest OOD score, and the reported SCOD risk is

\[\operatorname{SCOD\text{-}Risk@Cov}(\gamma) = r\left(\frac{k_\gamma}{N}\right) = \frac{1}{k_\gamma} \sum_{j=1}^{k_\gamma}\ell_{\sigma(j)}.\]

As input to forward and update, the metric accepts:

ood_scores (Tensor): Float tensor containing OOD scores, where larger values indicate more OOD-like samples.
classification_errors (Tensor): Boolean or binary tensor indicating misclassified ID samples.
is_ood (Tensor): Boolean or binary tensor indicating OOD samples.

As output to forward and compute, the metric returns:

scod_risk (Tensor): Scalar tensor containing the SCOD risk at the requested coverage. Lower values are better.

Parameters:

cov_threshold (float) – Target coverage \(\gamma\).
ood_cost (float) – Relative cost \(c_{\mathrm{OOD}}\) of accepting an OOD sample. The cost of accepting a misclassified ID sample is 1 - ood_cost. Defaults to 0.5.
kwargs – Additional keyword arguments passed to torchmetrics.Metric.

References