glmnet

Web: https://glmnet.stanford.edu/.

4.1-3

The ‘install.packages("glmnet")' command under R 4.1.3 gave the following error,

elnet_exp.cpp:141:59:   required from here
glmnetpp/include/glmnetpp_bits/elnet_point/gaussian_base.hpp:90:39: error: 'self' was not declared in this scope
                     [=](auto k) { self().template update<update_type::partial>(k, ab, dem); },

Bash

We now proceed manually,

Rscript -e 'download.packages("glmnet",".")'
tar tvfz glmnet_4.1-3.tar.gz

and then modify src/glmnetpp/include/glmnetpp_bits/elnet_point/gaussian_base.hpp line 90¹²³ as follows,

                     [=](auto k) {this -> self().template update<update_type::partial>(k, ab, dem); }

and similarly line 55.

Our final step is then

R CMD INSTALL glmnet

4.1-4

login node

It requires C++14, so we proceed with

echo "CXX14 = g++ -std=gnu++14 -fPIC" > ~/.R/Makevars
module load gcc/7
Rscript -e 'install.packages("glmnet")'

4.1-7

Released on 23/3/2023, it requires C++17 so the Makevars as above becomes

CXX17 = g++ -std=gnu++17 -fPIC

4.1-8

Besides ~/.R/Makevars, the following is necessary

module switch gcc/7

icelake

Latest information

After software update on 27/4/2022, the R 4.2.0 installed from login nodes also works nicely with glmnet installed there.

R/4.1.0-icelake

The Matrix package is also required to recompile.

module load R/4.1.0-icelake
wget https://cran.r-project.org/src/contrib/Matrix_1.4-1.tar.gz
wget https://cran.r-project.org/src/contrib/glmnet_4.1-4.tar.gz
R CMD INSTALL Matrix_1.4-1.tar.gz
R CMD INSTALL glmnet_4.1-4.tar.gz
R CMD INSTALL Matrix_1.4-1.tar.gz -l .

so that the Matrix package is installed first to get going with glmnet but then made available locally to avoid conflict with the login nodes.

library(Matrix,lib.loc=".")
library(glmnet)

The Matrix package is then reinstalled from the usual login node.