chagaz / celer

Fast solver for L1-type problems: Lasso, sparse Logisitic regression, Group Lasso, weighted Lasso, Multitask Lasso, etc.

celer

Fast algorithm to solve Lasso-like problems with dual extrapolation. Currently, the package handles the following problems: Lasso, Sparse Logistic regression, Group Lasso and Multitask Lasso. The estimators follow the scikit-learn API, come with automated cross-validation, and support sparse and dense data with feature centering and normalization. The solvers used allow for solving large scale problems with millions of features.

Documentation

Please visit https://mathurinm.github.io/celer/ for the latest version of the documentation.

Install the released version

Assuming you have a working Python environment, e.g. with Anaconda you can install celer with pip.

From a console or terminal install celer with pip:

pip install -U celer

Install and work with the development version

From a console or terminal clone the repository and install Celer:

git clone https://github.com/mathurinm/celer.git
cd celer/
pip install -e .

To build the documentation you will need to run:

pip install -U sphinx_gallery sphinx_bootstrap_theme
cd doc/
make html

Demos & Examples

You find on the documentation examples on the Leukemia dataset (comparison with scikit-learn) and on the Finance/log1p dataset (more significant, but it takes times to download the data, preprocess it, and compute the path).

Dependencies

All dependencies are in ./setup.py file.

Cite

If you use this code, please cite:

@InProceedings{pmlr-v80-massias18a,
  title =    {Celer: a Fast Solver for the Lasso with Dual Extrapolation},
  author =   {Massias, Mathurin and Gramfort, Alexandre and Salmon, Joseph},
  booktitle =        {Proceedings of the 35th International Conference on Machine Learning},
  pages =    {3321--3330},
  year =     {2018},
  volume =   {80},
}

@article{massias2019dual,
title={Dual Extrapolation for Sparse Generalized Linear Models},
author={Massias, Mathurin and Vaiter, Samuel and Gramfort, Alexandre and Salmon, Joseph},
journal={arXiv preprint arXiv:1907.05830},
year={2019}
}

ArXiv links:

• https://arxiv.org/abs/1802.07481
• https://arxiv.org/abs/1907.05830