Changelog#

v0.2.0 (2026-04-13)#

  • Add CV+ Mondrian conformal prediction sets for classification (binary and multiclass)

  • Add Mondrian-binned conditional prediction intervals for regression

  • New conformal subpackage with MondrianClassifierConformal and MondrianRegressorConformal

  • New NestedCVClassifier parameters: conformal_prediction, conformal_alpha

  • New NestedCVRegressor parameters: mondrian_bins, mondrian_min_bin_size

  • New ClassifierResults attributes: conformal_coverage_, conformal_set_size_stats_, conformal_qhat_per_fold_, conformal_report()

  • New RegressorResults attribute: mondrian_coverage_per_bin_

  • Fix RuntimeWarning in single-fold summary statistics

v0.1.1 (2026-03-09)#

  • Standardize numerical epsilon constants across the codebase

  • Fix prediction interval lower-quantile edge case

  • Fix Nadeau-Bengio t-test for zero-variance differences

  • Correct docstring parameter names and type references in plotting module

v0.1.0 (2026-03-06)#

Initial release.

  • Nested cross-validation for classification and regression

  • Post-hoc probability calibration (Platt, isotonic, beta, Venn-ABERS)

  • Threshold optimization (Youden’s J, F-beta, cost-sensitive, balanced accuracy, precision at recall)

  • Statistical model comparison (Nadeau-Bengio corrected t-test, Bayesian correlated t-test)

  • Hyperparameter stability diagnostics

  • Feature importance aggregation with Nogueira stability index

  • Callback system (progress, checkpointing, logging)

  • 25+ plotting functions

  • Full scikit-learn API compatibility