[R-sig-ME] Parallel version of lmer or glmer?

Pantelis Z. Hadjipantelis kalakouentin at gmail.com
Fri Aug 21 03:23:31 CEST 2015

There is no natively parallel versions of 'lmer' or 'glmer' to my knowledge.

That being said, if the BLAS implementation used (in this case 'Eigen' 
via 'RcppEigen') has multi-core capabilities one can use multiple cores.
Assuming one has OpenMP installed he would need to recompile 'lme4' 
changing the 'Makevars' arguments manually to activate OpenMP. I would 
not do this because I do not know if 'lme4' has been tested against 
multi-core installations; in theory this should make 'lme4' multi-core 
out-of-the-box though.
A starting point for more information on multi-threaded Eigen is here: 
http://eigen.tuxfamily.org/dox/TopicMultiThreading.html; prof. Bates can 
provide a more authoritative opinion on this matter.

One might want to try something safer/easier by changing the BLAS 
implementation R uses.
The BLAS implementation 'OpenBLAS' is relatively easy to install and 
provides a significant speed-up against the BLAS R is distributed with. 
I have used this configuration (and seen it in other people's systems) 
for years without any problems. This will not affect only 'lme4' but all 
the installed packages. In this way one accelerates (and makes 
multi-core) all the BLAS related functionality of R.
For more details on this just google: "openblas R"; they are a few 
guides online describing how to make this change.

Finally, if one is mostly interested in parallelism for the sake of 
bootstrapping, the function 'bootMer' allows bootstrapping a model in 
parallel directly (check the documentation of it  - '?lme4::bootMer').

All best,

On 08/20/2015 04:33 PM, María del Carmen Romero wrote:
> Hello,
> I want to know if there is a parallel version of lmer or glmer (both of
> package lme4).
> Thanks
> María
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

More information about the R-sig-mixed-models mailing list