Module calibration 

Model calibration for confidence estimation.

Calibration ensures model confidence matches actual accuracy.

Methods

• Temperature Scaling: Single parameter post-hoc calibration
• Platt Scaling: Logistic regression on logits
• Expected Calibration Error (ECE): Calibration metric

Structs

IsotonicRegression

Isotonic Regression calibrator.

PlattScaling

Platt scaling: fits logistic regression on logits.

TemperatureScaling

Temperature scaling calibrator. Divides logits by learned temperature T before softmax.

Functions

brier_score

Brier score - measures calibration quality.

expected_calibration_error

Expected Calibration Error (ECE) - measures calibration quality.

maximum_calibration_error

Maximum Calibration Error (MCE).

reliability_diagram

Generate reliability diagram data.