UniLasso: Univariate-Guided Sparse Regression

UniLasso is a Python package that implements a novel and interpretable approach to sparse regression corresponding to Univariate-Guided Sparse Regression (https://hdsr.mitpress.mit.edu/pub/3i97j340/release/4).

Features

• Implements UniLasso algorithm for various regression types (Gaussian, Binomial, Cox)
• Supports cross-validation for optimal regularization parameter selection
• Provides both fit and predict functionalities
• Includes utilities for data simulation

Installation

You can install UniLasso by running the following script.

git clone https://github.com/sophial05/uni-lasso.git
cd uni-lasso
pip install -e .

Quick Start

Here's a simple example of how to use UniLasso:

import numpy as np
from unilasso import *

# Generate some example data
X, y = simulate_gaussian_data(n = 100, p = 5)

# Fit UniLasso model using default lambda path
result = fit_unilasso(X, y, family='gaussian', lmdas = None)
plot_coef_path(result)

# Run UniLasso model using CV and plot CV error curve
cv_fit = cv_unilasso(X, y, family='gaussian')
plot_cv(cv_fit)
# extract the best cv model to use for prediction
extracted_fit = extract_cv(cv_fit)

# Generate some test data
X_test, y_test = simulate_gaussian_data(n = 100, p = 5)
# predict on test data
y_pred = predict(extracted_fit, X_test)

# Print coefficients
print("UniLasso coefficients:")
print(extracted_fit.coefs)

For a comprehensive examples on package usage, please refer to examples/ subdirectory.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use UniLasso in your research, please cite:

@article{chatterjee2025univariate,
  title={Univariate-Guided Sparse Regression},
  author={Chatterjee, Sourav and Hastie, Trevor and Tibshirani, Robert},
  journal={arXiv preprint arXiv:2501.18360},
  year={2025}
}

Contact

For any questions or feedback, please contact Sophia Lu at sophialu@stanford.edu