I have a Lanscos implementation of a rank-reduced SVD that can be distributed and handle very sparse matrixes with hundreds of millions of rows and columns. As you can imagine, it sometimes takes a few days to do its work.
I recently came across the Facebook fast randomized SVD at https://research.fb.com/fast-randomized-svd/ with code at http://tygert.com/software.html and I would like to try it, but the fortran code only supports dense matrix input, and at my scale there is no way I can use this.
So, does anyone know of a (preferably a distributed fortran) version of this or similar code that I could use?
Or, since the code I found includes source code, is there a simple way I can create a sparse matrix version from the dense matrix code?
FWIW, the code is built on BLAS and LAPACK.
Thanks for any help you can provide!
Aucun commentaire:
Enregistrer un commentaire