ConformalPredictor

conformal.ConformalPredictor(self, regressor, alpha=0.05)

A class used to represent a (Split) Conformal Predictor.

\[ \hat{C}_{\alpha}(x) = [\hat{u}(x) \pm q_{1 - \alpha}(\mathcal{S})] \]

Parameters

Name	Type	Description	Default
regressor	object	A regressor object that has a ‘predict’ method.	required
alpha	float	The significance level used in the prediction interval calculation.	0.05

Attributes

Name	Type	Description
scores	ndarray	The conformity scores of the calibration data.
quantile	float	The \((1 - \alpha)\) empirical quantile of the conformity scores.

Methods

Name	Description
fit	Calibrates the conformal predictor using the provided calibration set.
predict	Predicts the output for the given input X and provides a prediction interval.

fit

conformal.ConformalPredictor.fit(X, y)

Calibrates the conformal predictor using the provided calibration set.

Specifically, the fit method learns

\[ q_{1 - \alpha}(S) \]

where \(q_{1 - \alpha}(S)\) is the \((1 - \alpha)\) empirical quantile of the conformity scores

\[ \mathcal{S} = \{|y_i - \hat{u}(x_i)|\} \cup \{ \infty \} \]

Parameters

Name	Type	Description	Default
X	ndarray	The input data for calibration.	required
y	ndarray	The output data for calibration.	required

predict

conformal.ConformalPredictor.predict(X)

Predicts the output for the given input X and provides a prediction interval.

\[ \hat{C}_{\alpha}(x) = [\hat{u}(x) \pm q_{1 - \alpha}(\mathcal{S})] \]

Parameters

Name	Type	Description	Default
X	ndarray	The input data for which to predict the output.	required

Returns

Type	Description
tuple	A tuple containing the prediction (1D ndarray) and the lower (1D ndarray) and upper bounds (1D ndarray) of the prediction interval.