Stratified linear regression mixture models

Main classes

• Joint2Regressor: Stratified multiple regression with continuity constraints for two (or K) groups.
• JointKRegressor: Piecewise regression across K groups, with joint constraints, lasso and quantile.
• Joint2GMMRegressor: EM algorithm for piecewise Gaussian mixture regression with constraints.
• JointUtils: Utilities for group splitting, closest point to median finding, etc.

Key Features

• Joint multi-group regression with continuity or custom constraints at the join point.
• Supports quantile regression, penalized regression (lasso, ridge, elasticnet), and stratified GMM.
• Stratified multivariate piecewise regression, not directly available in scikit-learn or statsmodels.

Note : This package is provided “as is” for reproducing results, even if not all features are fully
implemented or tested. For the moment the lasso and ridge includes the $\beta_0$, hence the eventual
intercept may be removed (centering $y_\ell$). The code development has involved the use of modern
tools, for code refactoring into classes, docstring writing, code cleaning, debugging.

Usage

pip install stratifreg

import stratifreg

from stratifreg import Joint2Regressor

Datasets

	Name	n	p	X	y
D1	covid-19	4361	6	.csv	.csv
D2	pre-diabet	3059	4	.csv	.csv
D3	life-expectancy	2928	16	.csv	.csv
D4	pisa-2009	5233	20	.csv	.csv
D5	housing	20640	8	.csv	.csv

To be added next

• Smoother multi-points at joints
• Optimal computation for the $x_o$
• Statistical tests and variance
• Model and variable selection
• Full numpy implementation

References

• Priam, R. (2025). Family of linear regression mixture models stratified along the outcome. hal-04179813v3