On 2/19/11 9:20 AM, "Douglas Bates" <ba...@stat.wisc.edu> wrote:
>On Fri, Feb 18, 2011 at 4:50 PM, <ken.willi...@thomsonreuters.com> wrote: > >> Hi Doug, did you ever get a test case working for this? > >I didn't try further. I was hoping that you would pick up the sources >for the package and experiment with it. :-) Yes, a reasonable thing to hope for - I did pick it up and play with it some that day, but I didn't get very far. I messed with it again now and got farther, so now I'm getting correct output from it. BTW, I changed the prereq on Rcpp from 0.9.0 to 0.8.6 since that's the latest public release. > I was using a value of 712 for the number of eigenvalue-eigenvector > pairs to determine, which may have been a reason that it was taking so > long, relative to the example you show below where only 3 pairs are > determined. Looks like that was probably the case. Running my example matrix with your wrapper gives me the results I expect, in a fraction of a second. So now I'll modify your package to add return values for the eigenvectors, as well as the eigenvalues you returned, so that it looks like the return values from svd(). So it looks relatively straightforward (now) to package this up. I'm willing to do that, but would it be better to integrate this into the Matrix package, or something like that, so that it "just works" for a greater percentage of people? I asked Doug Rohde if his licensing terms would allow redistribution, and he said "SVDLIBC currently has an 'I don't care, use it for whatever' license. :-)" But since it's based in turn on SVDPACKC, it would be good to also contact its author, Michael Berry. > In my opinion Tim's code is > the state of the art in sparse matrix methods (except for the fact > that it is primarily in C rather than expressed as classes in C++). I may be missing it, but I don't see any SVD/eigenvector stuff in Tim's code. Is it an easy consequence of some other capability? -- Ken Williams Senior Research Scientist Thomson Reuters Phone: 651-848-7712 ken.willi...@thomsonreuters.com http://labs.thomsonreuters.com _______________________________________________ Rcpp-devel mailing list Rcpp-devel@lists.r-forge.r-project.org https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
