Enum ProbabilityCalibrationMethod

Namespace

AiDotNet.Enums

Assembly

AiDotNet.dll

Defines probability calibration strategies for classification-like outputs.

public enum ProbabilityCalibrationMethod

Fields

Auto = 0

Automatically selects a suitable calibration method based on the task and output shape.

IsotonicRegression = 4

Uses isotonic regression calibration (non-parametric monotonic calibration, typically for binary classification).

None = 1

Disables probability calibration.

PlattScaling = 3

Uses Platt scaling (logistic calibration, typically best for binary classification).

TemperatureScaling = 2

Uses temperature scaling (typically best for multiclass neural network probabilities/logits).

Remarks

Calibration transforms predicted probabilities to better reflect empirical correctness likelihoods.

For Beginners: Calibration helps ensure that "80% confident" means "correct about 80% of the time".