NEWS

---

probcal 0.1.0

• CRAN submission version.

probcal 0.0.0.9000

• Added multiclass calibration. cal_temperature() and cal_cv() accept a logit or probability matrix, and new constructors cal_vector_scaling(), cal_dirichlet(), and cal_ovr() cover vector scaling, Dirichlet calibration, and one-vs-rest calibration.
• ece(), mce(), ace(), and reliability_diagram() accept a probability matrix with a type argument for classwise or top-label confidence evaluation.
• Added mmce(), a binning-free Maximum Mean Calibration Error metric for binary and multiclass predictions.
• Added inst/CITATION so users can cite the package with citation("probcal").
• Added applied workflow, calibrator selection, and numerical validation vignettes.
• print() and summary() respect options(probcal.emoji = FALSE) to suppress the decorative glyph in console output.
• reliability_diagram() now reports ECE in the subtitle by default and can use either count-scaled or fixed-size points.
• Added optional development validation tests against Python netcal and R betacal.
• Initial development version with binary calibration methods, calibration metrics, reliability diagrams, and out-of-fold calibration support.