With 8+ core machines with 8GB or more memory becoming cheaper and more common, I've found that a decent solution for middle sized problems (i.e. too big for one processor, too small for m/r to help) is to just use a multi-threaded matrix/vector multiply.
I have a (somewhat hackish) subclass of SparseMatrix that does this for some SVDs we're running at work. If it's of general interest, I can try to work it into something more robust on our next hackday. -Jon On Mon, Oct 24, 2011 at 12:01 PM, Jake Mannix <jake.man...@gmail.com> wrote: > On Mon, Oct 24, 2011 at 11:49 AM, Sebastian Schelter < > ssc.o...@googlemail.com> wrote: > > > PageRank and RandomWalkWithRestart also rely on the problematic > iterative > > matrix vector multiplications... > > > > I wouldn't necessarily call these matrix-vector-multiplications > "problematic", but yes, > these are other good examples. > > -jake >
