linlearn: linear methods in Python

LinLearn is scikit-learn compatible python package for machine learning with linear methods. It includes in particular alternative "strategies" for robust training, including median-of-means for classification and regression.

Documentation | Reproduce experiments |

LinLearn simply stands for linear learning. It is a small scikit-learn compatible python package for linear learning with Python. It provides :

• Several strategies, including empirical risk minimization (which is the standard approach), median-of-means for robust regression and classification
• Several loss functions easily accessible from a single class (BinaryClassifier for classification and Regressor for regression)
• Several penalization functions, including standard L1, ridge and elastic-net, but also total-variation, slope, weighted L1, among many others
• All algorithms can use early stopping strategies during training
• Supports dense and sparse format, and includes fast solvers for large sparse datasets (using state-of-the-art stochastic optimization algorithms)
• It is accelerated thanks to numba, leading to a very concise, small, but very fast library

Installation

The easiest way to install linlearn is using pip

pip install linlearn

But you can also use the latest development from github directly with

pip install git+https://github.com/linlearn/linlearn.git

References