[R] Advice on use of R for Generalised Linear Modelling

Steve Lianoglou lianoglou.steve at gene.com
Sun Aug 11 22:48:57 CEST 2013


Hi,

On Sun, Aug 11, 2013 at 4:47 AM, Alan Sausse <alansausse at gmail.com> wrote:
> Hi,
>
> Not an expert R user, something of a novice - please be gentle with me!
>
> I have a particular interest in generalised linear models (GLMs) and I'm
> experienced in fitting them using other bits of software.
>
> R can fit GLMs of course, using the glm() command.  I have some large
> multivariate data sets I'd like to fit GLMs to, ideally using R.  Two
> concerns though:
>
> Firstly, I'm told that R isn't especially fast at fitting GLMs, especially
> if the data files are too large to fit into RAM.  Can anyone advise if
> there are alternatives to glm() around which might cope better.  For
> example, I've heard that RevolutionR is available, and claims to fit GLMs
> faster in these cases.  Might it be possible, alternatively, to write some
> very quick code using C (for example) and to get R to invoke this instead?
>  Has anyone tried to do this?

Likely not -- you'll need to have RevolutionR around for that, and if
you've have RevoR, then just use RevoR -- not sure what the point
would be call RevoR-specific functionality from R.

Perhaps the biglm package can help you from R, though, as it provides
a bigglm function that can do GLMs with out-of-memory data -- no idea
how well/fast it works, though.

You should also consider that your data may not require that, though
-- glmnet, for instance, works incredibly fast on large data. If your
data can actually be loaded (perhaps via a sparse matrix), then you
can try that.

HTH,
-steve

-- 
Steve Lianoglou
Computational Biologist
Bioinformatics and Computational Biology
Genentech



More information about the R-help mailing list