Asgl: A Python Package for Penalized Linear and Quantile Regression

TL;DR

Asgl is a Python package that simplifies solving penalized linear and quantile regression models, especially the adaptive sparse group lasso, using convex optimization and parallel computing for efficiency.

Contribution

It introduces a user-friendly Python package for penalized regression, emphasizing the adaptive sparse group lasso with optimized computation through multiprocessing.

Findings

Efficient implementation of penalized regression models in Python.

Supports high and low dimensional frameworks.

Reduces computation time via parallel processing.

Abstract

Asg is a Python package that solves penalized linear regression and quantile regression models for simultaneous variable selection and prediction, for both high and low dimensional frameworks. It makes very easy to set up and solve different types of lasso-based penalizations among which the asgl (adaptive sparse group lasso, that gives name to the package) is remarked. This package is built on top of cvxpy, a Python-embedded modeling language for convex optimization problems and makes extensive use of multiprocessing, a Python module for parallel computing that significantly reduces computation times of asgl.