v1.3.0 by @eugeneberta:
- Splitted calibrators.py into several subfiles and implemented many new calibrators,
most of which are benchmarked and ranked in the CalArena leaderboard.
Among others, added Binary histogram binning, Scaling-binning (from
uncertainty-calibration), BBQ (from netcal), Beta calibration (from betacal), ENIR
(from netcal), binary and multiclass Kernel calibration (using beta and dirichlet
kernels from ece_kde), Spline calibration (from splinecalib), CDF-Spline calibration,
Ensemble Temperature Scaling, tree based calibration with CatBoost, LightGBM and
XGBoost. - Re-structured the base
Calibratorclass to differentiate
_predict_proba_torch_implfrom_predict_proba_impl. - Added Kuiper and Kolmogorov-Smirnov binary calibration metrics.
- Deprecated python 3.9, added python 3.13 and 3.14 support.
What's Changed
- probmetrics v1.3.0 by @eugeneberta in #5
Full Changelog: v1.2.0...v1.3.0