High-dimensional penalized regression in python: the asgl library

A. Mendez-Civieta, M. C. Aguilera-Morillo, R. E. Lillo

High-dimensional big data requires robust variable selection to prevent overfitting. In this work we present asgl, a python package that offers a versatile framework for fitting penalized linear, logistic, and quantile regression models. Standard penalizations can yield biased estimates, whereas adaptive variants use specific weights to improve consistency. However, computing these weights is challenging in high-dimensional settings since standard unpenalized estimates fail. asgl addresses this methodological gap by providing built-in techniques (e.g., PCA, PLS, Ridge) for automated adaptive weight calculation. This feature makes complex models like the adaptive sparse group lasso viable for high-dimensional analysis. The library is fully integrated with the main ML framework in python, the scikit-learn ecosystem, allowing practitioners to seamlessly incorporate these penalizations into their analysis.

Keywords: High-dimension, Python, Regression, Classification, Variable-selection, Adaptive

GT TABiDa I
September 2, 2026  11:20 AM
Aula 24

S. Benítez Peña, S. Novo Díaz, J. García González Redondo

S. Bianconcini, S. Cagnone, L. Guastadisegni

P. Contró Rodríguez, S. Pineda Sanjuan, C. Ordaz Solera, A. Calviño Martínez