[Rcpp-devel] CppBugs vs WinBugs

[Watson, Samuel]

I am currently using CppBugs with Rcpp through R. I am very interested
to use CppBugs as I am finding WinBugs to be prohibitively slow, I use
large amounts of data in large multilevel models, so when I found
cppbugs I was excited. It says on the Github page for cppbugs that I can
achieve speeds of 20-100x faster than winbugs.

I have been testing some examples: for the linear model example that is
provided at Github, the time difference is about 2x

I also tested the Radon model with Rcpp and Andrew Gelmans data and for
10000 iterations and 10000 burnin with a thinning parameter of 5 the
difference is only 16 seconds vs 13 seconds.

 

I can run multiple chains in parallel through R with the 'snow' package,
cxxfunction() and package.skeleton() and this does seem to provide a
little bit of a boost (parallel winbugs vs parallel cppbugs). But
nothing greater than 4x.

 

Is it possible to achieve the kind of difference mentioned on Whit
Armstrong's page for cppbugs with R?

 

I am happy to post all the code if required.

 

Many thanks

Sam Watson

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/attachments/20111219/ab8d0b31/attachment.htm>