cvglmnetR.m

Simple replacement for the 'cvglmnet' function from the "glmnet in MATLAB" package that uses the R implementation of glmnet. Under the hood, a new R process is run each time a fit is performed. This is meant as a workaround for some bugs specific to the MATLAB version of the package. It can also be used to access newer features that have only been implemented for the R version of the package.

Although some effort has been made to make this into something fairly generic, cvglmnetR has been only tested with the 'binomial' and 'gaussian' family of models, under Linux, Mac OS and Windows.

Requirements

Besides MATLAB with "glmnet in MATLAB", you will need a recent R distribution with glmnet and R.matlab installed. doParallel is optional, and only necessary to make use of the parallel features of glmnet. These packages can be installed by the following command at the R prompt:

install.packages(c('glmnet', 'R.matlab', 'doParallel'))

Also, make sure that R is on your PATH. On Linux (and possibly Mac OS), this should not require any intervention, but sometimes there can be problems if you usually rely on adding custom folders to your PATH using .profile and similar. On Windows, it seems that by default R is not placed on the path when it is installed, so you will need to do something about it.

One way of ensuring the PATH is set as intended is to run something like the following command from the matlab prompt:

Linux/Mac OS:

setenv('PATH', ['/path/to/R/bin:', getenv('PATH')]);

Windows:

setenv('PATH', ['C:\Program Files\R\R-version\bin;', getenv('PATH')])

Taking care to substitute the actual path of your R bin folder instead of /path/to/bin or C:\Program Files\R\R-version\bin above.

Usage

The cvglmnetR function accepts the same arguments as cvglmnet from "glmnet in MATLAB" except keep and grouped, and returns a structure with the same fields as that returned by cvglmnet. Functions such as cvglmnetPredict and cvglmnetPlot should work when given this structure as the cvfit argument.

Note that if the parallel option is true, a parallel backend is automatically registered by calling registerDoParallel().

See example.m for some basic usage examples.

For further details, see the documentation for cvglmnet.

Other functionality

Partial, experimental support is also present for the following other packages:

• gglasso
• SGL

The interface for these packages is meant to be analogous to that for glmnet. This hasn't been sufficiently tested though, and is not described more in detail on purpose, to discourage careless use.

Licensing

This program is licensed under version 3 of the GPL or any later version. See COPYING for details.

Contributing

Bug reports and pull requests are welcome.