probly.conformal_prediction.methods.mondrian

Mondrian and Class-Conditional Conformal Prediction.

Classes

GroupedConformalBase(model, group_func)	Base class for group based conformal prediction (Mondrian & Class-Conditional).
MondrianConformalClassifier(model, score, ...)	Mondrian (region-wise) conformal predictor for classification.
MondrianConformalRegressor(model, score, ...)	Mondrian (region-wise) conformal predictor for regression.

class probly.conformal_prediction.methods.mondrian.GroupedConformalBase(model, group_func)

Bases: ConformalPredictor

Base class for group based conformal prediction (Mondrian & Class-Conditional).

Groups can be regions (Mondrian) or classes (ClassConditional). Shared calibrate() logic; subclasses implement predict().

Parameters:

• model (Predictor)

• group_func (ClassFunc)

static to_numpy(data)

Convert tensor or array-like to numpy array.

Parameters:

data (Any)

Return type:

ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)

Calibrate group-wise thresholds on a calibration dataset.

Parameters:

• x_cal (Sequence[Any])

• y_cal (Sequence[Any])

• alpha (float)

Return type:

float

group_func: ClassFunc

group_thresholds: dict[int, float | np.floating]

group_thresholds_lower: dict[int, float | np.floating]

group_thresholds_upper: dict[int, float | np.floating]

is_asymmetric: bool

score: Score

class probly.conformal_prediction.methods.mondrian.MondrianConformalClassifier(model, score, region_func, use_accretive=False)

Bases: GroupedConformalBase, ConformalClassifier

Mondrian (region-wise) conformal predictor for classification.

Parameters:

• model (Predictor)

• score (ClassificationScore)

• region_func (RegionFunc)

• use_accretive (bool)

static to_numpy(data)

Convert tensor or array-like to numpy array.

Parameters:

data (Any)

Return type:

ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)

Calibrate group-wise thresholds on a calibration dataset.

Parameters:

• x_cal (Sequence[Any])

• y_cal (Sequence[Any])

• alpha (float)

Return type:

float

predict(x_test, alpha, probs=None)

Return Mondrian prediction sets.

Parameters:

• x_test (Sequence[Any])

• alpha (float)

• probs (ndarray[Any, dtype[floating]] | None)

Return type:

ndarray[Any, dtype[bool]]

group_func: ClassFunc

group_thresholds: dict[int, float | np.floating]

group_thresholds_lower: dict[int, float | np.floating]

group_thresholds_upper: dict[int, float | np.floating]

is_asymmetric: bool

is_calibrated: bool

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

score: ClassificationScore

threshold: float | None

class probly.conformal_prediction.methods.mondrian.MondrianConformalRegressor(model, score, region_func)

Bases: GroupedConformalBase, ConformalRegressor

Mondrian (region-wise) conformal predictor for regression.

Parameters:

• model (Predictor)

• score (RegressionScore)

• region_func (RegionFunc)

static to_numpy(data)

Convert tensor or array-like to numpy array.

Parameters:

data (Any)

Return type:

ndarray[Any, dtype[floating]]

calibrate(x_cal, y_cal, alpha)

Calibrate group-wise thresholds on a calibration dataset.

Parameters:

• x_cal (Sequence[Any])

• y_cal (Sequence[Any])

• alpha (float)

Return type:

float

predict(x_test, alpha)

Return Mondrian prediction intervals.

Parameters:

• x_test (Sequence[Any])

• alpha (float)

Return type:

ndarray[Any, dtype[floating]]

group_func: ClassFunc

group_thresholds: dict[int, float | np.floating]

group_thresholds_lower: dict[int, float | np.floating]

group_thresholds_upper: dict[int, float | np.floating]

is_asymmetric: bool

is_calibrated: bool

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

score: RegressionScore

threshold: float | None