Self-Calibrating Conformal¶

selfcalibratingconformal is a Python package for post-hoc calibration and conformal prediction with black-box regression models.

It supports two workflows:

SelfCalibratingConformalPredictor for calibrated point predictions, Venn-Abers style prediction sets, and adaptive prediction intervals
VennAbersQuantileConformalPredictor for conformal prediction based on a calibrated predictor of the (1 - alpha) quantile of a conformity score

In both workflows, interval width can adapt across data-adaptive bins learned by isotonic regression.

Installation¶

python -m pip install selfcalibratingconformal

For development:

python -m pip install -e ".[dev,docs]"

Quickstart¶

import numpy as np

from selfcalibratingconformal import SelfCalibratingConformalPredictor


class MeanPredictor:
    def predict(self, x):
        x = np.asarray(x)
        return 1.5 * x[:, 0]


model = SelfCalibratingConformalPredictor(MeanPredictor())
model.fit(X_cal, y_cal, alpha=0.1)

point_predictions = model.predict_point(X_test)
intervals = model.predict_interval(X_test)
coverage, width = model.check_coverage(X_test, y_test)

Workflows¶

Regression workflow¶

Use SelfCalibratingConformalPredictor when you already have a point predictor for y and want calibrated predictions and intervals.

Quantile workflow¶

Use VennAbersQuantileConformalPredictor when your workflow predicts the (1 - alpha) quantile of a conformity score and you want intervals built from a calibrated threshold.

Documentation¶

Papers¶

References¶

van der Laan, L. and Alaa, A. M. (2024). Self-Calibrating Conformal Prediction.
van der Laan, L. and Alaa, A. M. (2025). Generalized Venn and Venn-Abers Calibration with Applications in Conformal Prediction.
Vovk, V., Petej, I., and Fedorova, V. (2015). Large-scale probabilistic predictors with and without guarantees of validity.
Angelopoulos, A. N. and Bates, S. (2023). Conformal prediction: A gentle introduction.
Romano, Y., Patterson, E., and Candes, E. (2019). Conformalized quantile regression.