Ted Dunning wrote: > Also, it isn't entirely clear yet whether power iterations are more > efficient than simply increasing the fudge factor p. Power iterations are > very effective, and increasing p increases costs in the cube, but running > MR passes is expensive enough that some increase in p might be sufficient > and still faster than a power iteration.
Don't even think about it -- on large matrices (significant truncation), the randomized SVD without power iterations is complete rubbish, regardless of any sane value of `p`. (assuming your fudge factor p = oversampling parameter `p` from Halko et al.). You can see some accuracy experiments (wikipedia, images) here: http://groups.google.com/group/gensim/msg/1f3106b85837ce9c and here http://groups.google.com/group/gensim/msg/240a348b70f18b30 . Also see the end of that (longish) thread for some notes on accuracy in face of many power iterations (~ beware of numeric overflows). Best, Radim
