Calling R from Matlab

Update on 8 Jan 2011: A newer set of instructions can be found here. The notes below are somewhat out of date.

This uses the COM interface, so only works on Windows.

Install R.2.7.0 or newer. Make sure R.exe is in your windows path.
Install and run the Rcom server
Reboot matlab and put the following m-files in your path
Optional: in R-2.7.0\bin make a file called "R.init" which contains startup commands, such as
```
library(glasso)
```

In matlab, type Rdemo. This should display

b =
     1     4     9    16    25    36    49    64    81   100
c =
     2     5    10    17    26    37    50    65    82   101

You can also call R as a subprocess as follows

!R --no-save   # at the matlab prompt
q() #  at the R prompt to quit.

Trouble shooting

If it's not working make sure the Dcom service is running under Control Panel --> Administrative Tools --> Services --> DCOM Server Process Launcher. You can start the service by right clicking and selecting 'start', (unless of course its already running) and have it always start on reboot by double clicking and changing the startup type to automatic.
The Rsvr install creates a bunch of shortcuts under the R program folder on the start bar. One of these, "server 01 Basic Test" will test the connection to R - if that works we know the problem is between Matlab and Dcom and also unlikely to be a firewall problem.

Example

A matlab wrapper to the glasso function.

function precMat = glassoR(C, rho, useMBapprox)
% Use L1-linear regression to find markov blankets and then Lambda.
% Uses R code from http://www-stat.stanford.edu/~tibs/glasso/
openR;
evalR('C<-1') % must pre-declare variable before writing a matrix
evalR('L<-1') 
putRdata('C',C);
putRdata('rho',rho);
putRdata('useMBapprox', useMBapprox)
evalR('stuff <- glasso(C,rho=rho,approx=useMBapprox)'); 
evalR('L <- stuff$wi') % inverse covariance matrix
precMat = getRdata('L');
closeR;