There are many alternatives for L1-regularized generalized linear models in R, but none that utilizes the efficient SAGA algorithm despite its excellent convergence properties and track record of compelling empirical results. A successful implementation of the SAGA algorithm does, however, exist in scikit-learn: an actively maintained and well-documented Python module. The goal of this project is to port that implementation to R as a package targeted for submission to the Comprehensive R Archive Network (CRAN).The end result will be an easy-to-use and blazingly fast algorithm for L1-regularized models, which will come as a wanted addition to the toolkit of R users interested in big data modelling.


Johan Larsson


  • Michael Weylandt
  • Toby Hocking