Module types 

Auto-generated module

🤖 Generated with SplitRS

Structs

BayesianBinningQuantiles

Bayesian Binning into Quantiles (BBQ) - sophisticated histogram-based calibration Bins predictions into quantiles and learns Bayesian posterior for each bin Uses Beta distribution for robust probability estimation with uncertainty quantification

IsotonicRegression

Isotonic Regression - non-parametric calibration using monotonic transformation More flexible than Platt scaling but requires more data

MatrixScaler

Matrix Scaling - full affine transformation for maximum calibration flexibility Uses full weight matrix W and bias vector b: calibrated = softmax(W @ logits + b) More expressive than vector scaling but requires more data to avoid overfitting

PlattScaler

Platt Scaling - fits a logistic regression on decision scores Calibrates binary classifier outputs to produce better probability estimates

TemperatureScaler

Temperature Scaling - simple and effective multi-class calibration Scales logits by a single learned temperature parameter Particularly effective for neural network outputs

VectorScaler

Vector Scaling - extension of temperature scaling with class-specific parameters Uses diagonal weight matrix and bias vector for more flexible calibration Particularly effective when different classes have different calibration needs