Beyond L1: Faster and Better Sparse Models with skglm

TL;DR

This paper introduces a rapid algorithm for estimating large-scale sparse generalized linear models with various penalties, significantly outperforming existing methods in speed and flexibility.

Contribution

It presents a novel, scalable algorithm that efficiently handles complex sparse models, including non-convex penalties, with a user-friendly scikit-learn compatible implementation.

Findings

Solves models with millions of samples and features in seconds

Outperforms state-of-the-art algorithms in speed and accuracy

Supports customized data fits and penalties

Abstract

We propose a new fast algorithm to estimate any sparse generalized linear model with convex or non-convex separable penalties. Our algorithm is able to solve problems with millions of samples and features in seconds, by relying on coordinate descent, working sets and Anderson acceleration. It handles previously unaddressed models, and is extensively shown to improve state-of-art algorithms. We provide a flexible, scikit-learn compatible package, which easily handles customized datafits and penalties.