ProbabilityCalibrationMethod

Enums · AiDotNet.Enums

Defines probability calibration strategies for classification-like outputs.

For Beginners

Calibration helps ensure that “80% confident” means “correct about 80% of the time”.

How It Works

Calibration transforms predicted probabilities to better reflect empirical correctness likelihoods.

Fields

Field Summary
ProbabilityCalibrationMethod Auto Automatically selects a suitable calibration method based on the task and output shape.
ProbabilityCalibrationMethod BayesianBinning Bayesian Binning into Quantiles (BBQ) - adaptive binning.
ProbabilityCalibrationMethod BetaCalibration Uses beta calibration (more flexible than Platt scaling, handles asymmetric distortions).
ProbabilityCalibrationMethod HistogramBinning Histogram Binning - assigns average probability to each bin.
ProbabilityCalibrationMethod IsotonicRegression Uses isotonic regression calibration (non-parametric monotonic calibration, typically for binary classification).
ProbabilityCalibrationMethod None Disables probability calibration.
ProbabilityCalibrationMethod PlattScaling Uses Platt scaling (logistic calibration, typically best for binary classification).
ProbabilityCalibrationMethod TemperatureScaling Uses temperature scaling (typically best for multiclass neural network probabilities/logits).
ProbabilityCalibrationMethod VennABERS Venn-ABERS - provides probability intervals, not point estimates.
int value__