eval_toolkit.metrics

DEFAULT_ASSUMED_PRIORS	Built-in immutable sequence.
ThresholdResult	Outcome of operating-point selection at a given criterion.
brier_decomposition	Murphy 1973 [#murphy]_ decomposition of the Brier score.
brier_score
expected_calibration_error	Expected calibration error on equal-width probability bins.
expected_calibration_error_debiased	Bias-corrected L1 ECE via simulated-H0 Monte-Carlo (Roelofs 2022 spirit).
expected_calibration_error_equal_mass	ECE on equal-mass (quantile) bins.
expected_calibration_error_l2	Equal-mass L2 ECE — root mean squared bin-level miscalibration.
expected_calibration_error_l2_debiased	Bias-corrected L2 ECE per Kumar 2019 [#kumar]_ §3.3.
headline_metrics	Bundle PR-AUC + ROC-AUC + 3 operating-point F1s + per-stratum recall (if provided).
metrics_at_threshold	Precision / recall / F1 / accuracy / TN/FP/FN/TP at a fixed threshold.
pr_auc
precision_at_prior	Project precision under a different positive-class prior.
quantile_stratified_pr_auc	PR-AUC on the central [q_low, q_high] range of any 1-D stratifier.
quantile_stratified_report	Full vs trimmed PR-AUC report with a gap-flag (SDD reporting convention).
roc_auc
score_distribution_summary	Threshold-free score-distribution summary.
single_class_threshold_metrics	Operating metrics for all-positive or all-negative slices.
stratified_recall	Recall (TPR) per categorical stratum.