just for clarity -- both SVD implementations we are talking about are coming from lapack -- they are just different SVD algorithms present in lapack. Numpy chosen one and exposed it in its interface. Swaroop just added additional interface to get access to the other one in case lapack is installed. Theoretically they both should provide very similar, if not identical, results. It is the corner cases (ill conditioned etc) where I expect them to be different. also indeed it would be neat to know performance wise.
> I also have questions about singular value decomposition convergence > in hyperalignment and what it means when it does not converge, but > that is a topic for another thread... by default it is the first dataset which serves as the 'target' to kick off the hyperalignment... if SVD implementation happens to not converge, it tries to use the next dataset and so on... in an unfortunate case that none of them converged (I bet in case of really obscure data) you could try different SVD implementation for Procrustean or just look at WTF "interesting" with your data ;-) -- =------------------------------------------------------------------= Keep in touch www.onerussian.com Yaroslav Halchenko www.ohloh.net/accounts/yarikoptic _______________________________________________ Pkg-ExpPsy-PyMVPA mailing list [email protected] http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
