From: Tonio W. <ton...@un...> - 2006-07-17 09:50:20
|
Hello, well, congratulations to your daughter, Dominic! I can deeply imagine tha= t you are now occupied by more important things than software maintenance... Thanks for your advice concerning my questions. Indeed, there is a built-in limit to 3000, and I did not dare to modify it. However, I don't see why, (using Michael Berry's GTP) I applied SVD to much larger matrice= s (~120.000 cols). I will try that out and I'll let you know about my results. Best wishes, Tonio > > Hi Tonio, David and Dominic, > > On Sat, 15 Jul 2006, Dominic Widdows wrote: > >> I agree with the solutions you posted below, couldn't have done better >> myself. It's possible to set COLUMNS in the default-params file, I >> don't know of any flag that prevents the number from being greater tha= n >> 2999. Nor do I know off hand whether the matrix format in memory is >> scalable enough to handle a full term-by-term matrix. But this is >> certainly the place to begin looking. > > The SVD interface sets a limit on the number of columns: > #define NMAX 3000 /* bound on ncol, order of A */ > (it's in infomap-nlp/svd/svdinterface/las2.c) > > I don't know whether this line is part of the original SVDPACKC > package (which was developed at the University of Tennesse and not by t= he > infomap project). It probably is. You can try increasing the limit and > see if the singular value decomposition is still successful. > > Good luck! > Beate > > --=20 Wissenschaftlicher Mitarbeiter Institut f=FCr Kognitionswissenschaft Universit=E4t Osnabr=FCck Raum 31/450c Albrechtstra=DFe 28 49076 Osnabr=FCck Tel: 0541/969-3391 |