Struct CalibratedClassifierCV 

#[non_exhaustive]
pub struct CalibratedClassifierCV { /* private fields */ }

A calibrated classifier wrapper.

Uses internal cross-validation to produce calibrated probability estimates from any classifier that supports predict_proba.

During fit, the data is split into n_folds folds. For each fold, the base classifier is trained on the training portion, predictions are made on the held-out portion, and those predictions are used to fit a calibration model (per class in the OVR scheme). At predict time, the full model’s raw probabilities are transformed through the calibrator.

Example

use scry_learn::calibration::{CalibratedClassifierCV, CalibrationMethod};
use scry_learn::tree::DecisionTreeClassifier;
use scry_learn::dataset::Dataset;

let data = Dataset::from_csv("iris.csv", "species").unwrap();
let mut cal = CalibratedClassifierCV::new(
    DecisionTreeClassifier::new(),
    CalibrationMethod::Isotonic,
).n_folds(5);

cal.fit(&data).unwrap();
let probs = cal.predict_proba(&data.feature_matrix()).unwrap();

Implementations

impl CalibratedClassifierCV

pub fn new<C: CalibrableClassifier + 'static>( classifier: C, method: CalibrationMethod, ) -> Self

Create a new calibrated classifier wrapper.

pub fn n_folds(self, n: usize) -> Self

Set the number of cross-validation folds (default: 5).

pub fn fit(&mut self, data: &Dataset) -> Result<()>

Fit the calibrated classifier.

1. Splits data into n_folds stratified folds.
2. For each fold, trains a clone of the base classifier on the training portion and collects predict_proba on held-out.
3. Fits a per-class calibrators on the aggregated out-of-fold predictions.
4. Re-trains the base classifier on the full dataset.

pub fn predict_proba(&self, features: &[Vec<f64>]) -> Result<Vec<Vec<f64>>>

Predict calibrated probabilities.

Returns [n_samples][n_classes] with calibrated probabilities that sum to 1 for each sample.

pub fn predict(&self, features: &[Vec<f64>]) -> Result<Vec<f64>>

Predict class labels (argmax of calibrated probabilities).