emanuele at relativita.com
Fri Sep 19 21:36:19 UTC 2008
Per B. Sederberg wrote:
> I haven't tried them, but was looking for some existing implementation
> (so we didn't have to do all the work.) I've heard of people having
> good success with them (for example, the team that won the PBAIC one
> year, http://www.ebc.pitt.edu/2007/competition.html used them.)
RVM is an approximated (and sparse) Gaussian process :) with
(see pag.149 - or better 21).
The good thing of RVM is that they are very fast to compute.
I know quite well the team you mentioned and their work on RVM.
They are the SPM guys.
I hope to work on sparse GP soon. They should be fast enough
compared to RVM and without major drawbacks ;)
Anyway RVMs are interesting and they can use ARD kernels,
which is of highest interest for me/us.
As far as I know RVM should not be difficult to implement in
PyMVPA. And we have beautiful kernel.py already done :D
More news later.
More information about the Pkg-ExpPsy-PyMVPA