Module linfa::composing::platt_scaling

Implement Platt calibration with Newton method

Platt scaling is a way of transforming the outputs of a classification model into a probability distribution over classes. It is, for example, used in the calibration of SVM models to predict plausible probability values.

Example

let model = ...;

let model = Platt::params()
     .fit_with(model, &train)?;

let pred: Array1<Pr> = model.predict(&valid);

Structs

• Platt
  Fitted Platt model
• PlattParams
• PlattValidParams
  Parameters for Platt’s Newton method

Enums

• PlattError
  Platt Newton’s method errors

Functions

• platt_newton_method
  Run Newton’s method to find optimal A and B values
• platt_predict
  Predict a probability with the sigmoid function