ExclusiveLasso

Generalized Linear Models with the Exclusive Lasso Penalty

Authors

Affiliations

Michael Weylandt

Department of Statistics, Rice University

Frederick Campbell

Department of Statistics, Rice University

Genevera I. Allen

Departments of Statistics, Electrical & Computer Engineering, and Computer Science, Rice University

The ExclusiveLasso R package implements the Exclusive Lasso penalty as analyzed by Campbell and Allen (EJS, 2017), for Gaussian, logistic, and Poisson regression. The interface is glmnet compatible and the package supports both proximal gradient and coordinate descent solvers.

The package vignette is also notable for a thorough derivation of the back-tracking proximal Newton algorithm used for penalized GLMs, giving more detial than I have found elsewhere. Reading it would be useful for 1st or 2nd year Ph.D. students interested in efficient implementation of this family of algorithms.

Direct Link: http://github.com/DataSlingers/ExclusiveLasso

Package Documentation: https://DataSlingers.github.io/ExclusiveLasso/