Module classification 

Classification metrics — accuracy, the binary confusion matrix and its derived rates, ROC AUC (rank form, reusing the statistics ranker), and log-loss.

Structs

ConfusionBinary

Binary confusion matrix (positive class = true).

Functions

accuracy

Fraction of exact matches between predicted and true class labels. None if lengths differ or are empty.

confusion_binary

Build the binary confusion matrix from predicted/true boolean labels.

log_loss

Binary cross-entropy (log-loss): −(1/n)Σ[yᵢln pᵢ + (1−yᵢ)ln(1−pᵢ)], with p clamped away from 0/1 for numerical safety. None on a length mismatch.

roc_auc

ROC AUC via the Mann–Whitney rank statistic: AUC = (R₊ − n₊(n₊+1)/2) / (n₊·n₋), where R₊ is the sum of the (tie-averaged) ranks of the positive-class scores. Reuses the statistics ranker. None if the inputs mismatch or a class is empty.