mixlasso

This package extends IPFStructPenalty for stratum-specific feature selection by using varying coefficients and random effects. The core algorithms of tree lasso have been coded in C++ for computational efficiency.

R package instruction

You can download the package and install it locally, or install it from github directly by the code:

library(devtools)
devtools::install_github("zhizuio/mixlasso")

Troubleshooting

Using all cores on HPC clusters

When running this package without parallelization, it might occupy all cores on HPC clusters. This is because an R dependency mlegp with setting BLAS/LAPACK uses all available cores by default. One solution is to set BLAS/OpenMP environment variables to control allocation:

export MKL_NUM_THREADS=1
export NUMEXPR_NUM_THREADS=1
export OMP_NUM_THREADS=1