[Research] Thin SVD code in auton codebase?

[Paul Komarek]

Aha, here it is (author was munk): the name is propack:
http://soi.stanford.edu/~rmunk/PROPACK/

It's still one of the top 10 hits for "partial svd fortran", though
maybe there's less competition on that search than in the old days...

Paul

On Mon, Nov 16, 2009 at 10:18 AM, Paul Komarek <komarek.paul at gmail.com> wrote:
> I think I checked in a modified fortran version, that I had tried to
> make callable from C.  It seemed like there was some issue, but I
> don't recall for sure.  It was wicked-fast, and supported partial
> decomposition.  Unfortunately, I can't remember the name of it, but
> the author had "monk" in his name and there might be some f2c or other
> fortran files in the directory -- maybe play with grep.
>
> Paul
>
> On Mon, Nov 16, 2009 at 9:10 AM, Sajid Siddiqi <siddiqi at cs.cmu.edu> wrote:
>> Hi,
>>
>> Does anyone know if there is code for doing a thin SVD (i.e. top k
>> singular values and vectors) in the auton codebase?
>>
>> The svd.c code in /utils seems to return the full SVD component matrices
>> for an m x n matrix, i.e. a m x n matrix U, a  n x n diagonal matrix S (as
>> a dyv), and an n x n matrix V. If m and n are both large, these matrices
>> can be slow to compute. If only the top few singular values and vectors
>> are computed (i.e. the 'thin' SVD) it would be much faster.
>>
>> Thanks,
>> Sajid
>>
>> _______________________________________________
>> Research mailing list
>> Research at autonlab.org
>> https://www.autonlab.org/mailman/listinfo/research
>>
>