pitcp.models.CQR

class CQR(estimator, *, confidence_level=0.9, gamma=0.5)[source]

Fits conformalized quantile intervals with a boosting estimator.

The estimator can be a scikit-learn histogram gradient booster or an optional LightGBM, XGBoost, or CatBoost regressor. Backend-specific quantile models are fitted independently for every target output.

Quantile estimation settings:

  • estimator: Unfitted regression prototype cloned into lower- and upper-quantile estimators for every target output. Supported backends are histogram gradient boosting, LightGBM, XGBoost, and CatBoost.

  • confidence_level: Target marginal coverage probability in the open interval from zero to one. Defaults to 0.9.

  • gamma: Fraction of total miscoverage assigned to the lower tail, from zero to one. Defaults to 0.5, which gives equal-tailed intervals.

Variables:

  • estimator (BaseEstimator) – Unfitted quantile-regression prototype.

  • confidence_level (float) – Desired marginal coverage level.

  • gamma (float) – Miscoverage fraction assigned to the lower tail.

  • estimators (list[QuantileEstimator]) – Fitted quantile adapter for each target output.

  • scores (np.ndarray) – Joint calibration scores.

  • correction (float) – Finite-sample conformal correction.

Parameters:

  • estimator (BaseEstimator)

  • confidence_level (float)

  • gamma (float)

Examples

>>> from sklearn.ensemble import HistGradientBoostingRegressor
>>> from pitcp import CQR
>>> model = CQR(HistGradientBoostingRegressor(), confidence_level=0.9)
>>> model.fit([[0.0], [1.0], [2.0]], [0.0, 1.0, 2.0])
CQR(...)
>>> model.conformalize([[3.0], [4.0]], [3.0, 4.0])
CQR(...)
>>> model.predict([[5.0]]).shape
(1, 2)
__init__(estimator, *, confidence_level=0.9, gamma=0.5)[source]

Initializes the conformalized quantile estimator.

Parameters:

  • estimator (BaseEstimator) – HistGradientBoosting, LightGBM, XGBoost, or CatBoost regressor.

  • confidence_level (float, optional) – Coverage level. Defaults to 0.9.

  • gamma (float, optional) – Miscoverage assigned to the lower tail. Defaults to 0.5.

fit(X, y)[source]

Fits cloned quantile estimators.

Parameters:

  • X (np.typing.ArrayLike) – Training features with shape (n_samples, n_features).

  • y (np.typing.ArrayLike) – Targets with shape (n_samples,) or (n_samples, n_outputs).

Returns:

The fitted estimator.

Return type:

Self

Raises:

TypeError – If the estimator backend is unsupported.

conformalize(X, y)[source]

Calibrates a joint correction on held-out targets.

Parameters:

  • X (np.typing.ArrayLike) – Calibration features with shape (n_samples, n_features).

  • y (np.typing.ArrayLike) – Calibration targets with shape (n_samples,) or (n_samples, n_outputs).

Returns:

The calibrated estimator.

Return type:

Self

predict(X)[source]

Predicts calibrated lower and upper bounds.

Parameters:

X (np.typing.ArrayLike) – Features with shape (n_samples, n_features).

Returns:

Bounds with shape (n_samples, 2, n_outputs), with a

singleton output axis collapsed.

Return type:

np.ndarray

contains(X, y)[source]

Tests whether targets lie inside every output interval.

Parameters:

  • X (np.typing.ArrayLike) – Features with shape (n_samples, n_features).

  • y (np.typing.ArrayLike) – Targets with shape (n_samples,) or (n_samples, n_outputs).

Returns:

One boolean per sample.

Return type:

np.ndarray

set_fit_request()

No-op.

Calling this method has no effect.

Returns:

self – The updated object.

Return type:

object

Parameters:

self (CQR)

set_predict_request()

No-op.

Calling this method has no effect.

Returns:

self – The updated object.

Return type:

object

Parameters:

self (CQR)

set_score_request(*, sample_weight='$UNCHANGED$')

Configure whether metadata should be requested to be passed to the score method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config()). Please check the User Guide on how the routing mechanism works.

The options for each parameter are:

  • True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.

  • False: metadata is not requested and the meta-estimator will not pass it to score.

  • None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

  • str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

  • sample_weight (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for sample_weight parameter in score.

  • self (CQR)

Returns:

self – The updated object.

Return type:

object