Ali-Mahzarnia/222: Computes the solution path of the multivariate Scalar-on-Functional Elastic Net regression in serial and parallel

This R package runs the Group Elastic Net (including lasso, ridge, elastic net, and ordinary least square) regression with scalar response values and observed functional covariates. In addition, it penalizes the curvature of the output by implementing a penalty on the second derivative of the estimated coefficient curves. One of the two algorithms of this package is ADMM (mostly developed in C++). ADMM is designed for parallel computations and is only recommended on systems equipped with many strong cores. This algorithm runs parallel on Linux, but it runs serial on Windows. The second algorithm uses the fGMD package that is built exclusively for this package. The fGMD package is a heavily modified version of the gglasso package (BMD and Strong Rule). The features added to the original gglasso package are: the mixing parameter (alpha) and its net search cross-validation, the curvature penalization for functional regression and its net search cross-validation, the optimized Fortran core function to accelerate the curvature penalization updates, and the progress reports with time estimations. For this package to work, first install fGMD. The fGMD package does not work independently from this package, and it does not interfere with the functions of the gglasso package due to the slight name differences.

install.packages("remotes")
remotes::install_github("Ali-Mahzarnia/222")