probly.conformal_prediction.methods.split¶

Split conformal prediction methods.

Classes

`SplitConformal`([calibration_ratio, random_state])	Utility to split data into training and calibration sets.
`SplitConformalClassifier`(model, score[, ...])	Generic split conformal predictor for classification.
`SplitConformalPredictor`(model)	Generic split conformal predictor base class.
`SplitConformalRegressor`(model, score)	Generic split conformal predictor for regression.

class probly.conformal_prediction.methods.split.SplitConformal(calibration_ratio=0.3, random_state=None)[source]¶

Bases: object

Utility to split data into training and calibration sets.

Parameters:

calibration_ratio (float)
random_state (int | None)

split(x, y, calibration_ratio=None)[source]¶

Split data into training and calibration sets.

Parameters:

x (Sequence[Any])
y (Sequence[Any])
calibration_ratio (float | None)

Return type:

tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]

class probly.conformal_prediction.methods.split.SplitConformalClassifier(model, score, use_accretive=False)[source]¶

Bases: SplitConformalPredictor, ConformalClassifier

Generic split conformal predictor for classification.

Parameters:

model (Predictor)
score (ClassificationScore)
use_accretive (bool)

static to_numpy(x)¶

Convert tensor to NumPy on CPU (float dtype).

Parameters:: x (Any)
Return type:: ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)¶

Calibrate the predictor on a calibration dataset.

Parameters:

x_cal (Sequence[Any])
y_cal (Sequence[Any])
alpha (float)

Return type:

predict(x_test, alpha, probs=None)[source]¶

Return prediction sets as a (n_instances, n_labels) 0/1-matrix.

Parameters:

x_test (Sequence[Any])
alpha (float)
probs (Any)

Return type:

npt.NDArray[np.bool_]

score: ClassificationScore¶

class probly.conformal_prediction.methods.split.SplitConformalPredictor(model)[source]¶

Bases: ConformalPredictor

Generic split conformal predictor base class.

Parameters:: model (Predictor)

static to_numpy(x)[source]¶

Convert tensor to NumPy on CPU (float dtype).

Parameters:: x (Any)
Return type:: ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)[source]¶

Calibrate the predictor on a calibration dataset.

Parameters:

x_cal (Sequence[Any])
y_cal (Sequence[Any])
alpha (float)

Return type:

score: Score¶

class probly.conformal_prediction.methods.split.SplitConformalRegressor(model, score)[source]¶

Bases: SplitConformalPredictor, ConformalRegressor

Generic split conformal predictor for regression.

Parameters:

model (Predictor)
score (RegressionScore)

static to_numpy(x)¶

Convert tensor to NumPy on CPU (float dtype).

Parameters:: x (Any)
Return type:: ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)[source]¶

Calibrate thresholds for regression (supports symmetric and CQR).

Parameters:

x_cal (Sequence[Any])
y_cal (Sequence[Any])
alpha (float)

Return type:

predict(x_test, alpha)[source]¶

Return prediction intervals as a (n_instances, 2)-matrix [lower, upper].

Parameters:

x_test (Sequence[Any])
alpha (float)

Return type:

npt.NDArray[np.floating]

is_calibrated: bool¶

nonconformity_scores: npt.NDArray[np.floating] | None¶

score: RegressionScore¶

threshold: float | None¶