Probablistic Calibration with Regression Splines

scikit-learn compatible

pip install splinator

Supported models:

• Linear Spline Logistic Regression

Supported metrics:

• Spiegelhalter’s z statistic
• Expected Calibration Error (ECE)

[1] You can find more information in the Linear Spline Logistic Regression.

[2] Additional readings

• Zhang, Jian, and Yiming Yang. Probabilistic score estimation with piecewise logistic regression. Proceedings of the twenty-first international conference on Machine learning. 2004.
• Guo, Chuan, et al. "On calibration of modern neural networks." International conference on machine learning. PMLR, 2017.

comparison	notebook
scikit-learn's sigmoid and isotonic regression
pyGAM’s spline model

The dependencies are managed by pdm

To run tests, run pdm run -v pytest tests

from splinator.estimators import LinearSplineLogisticRegression
import numpy as np

# random synthetic dataset
n_samples = 100
rng = np.random.RandomState(0)
X = rng.normal(loc=100, size=(n_samples, 2))
y = np.random.randint(2, size=n_samples)

lslr = LinearSplineLogisticRegression(n_knots=10)
lslr.fit(X, y)