[SciPy-dev] more with SVD wrapping

M Trumpis mtrumpis@berkeley....
Mon Jun 8 17:31:09 CDT 2009

Hi all,

While I was playing with the f2py wrappings for SVD routines, I went
ahead and beefed up the *gesdd and *gesvd interfaces, and added a
wrapping for the "generalized SVD" wrappings, *ggsvd.

improvements to svd:
* can query the lapack drivers for best workspace size--this
eliminates the need for the code in "calc_lwork.f".
* can ask the drivers to return compact matrices: a = (m x n) ==> u =
(m x min(m,n)), vh = (min(m,n) x n)

Use of the *gesvd driver can be toggled with an argument:
scipy.linalg.svd( ... , fast_svd=False)

I made two scipy interpretations of the GSVD, one corresponding to the
lapack docs, and one that is more like the matlab version.

This ticket supercedes ticket 957.


Kind regards,

More information about the Scipy-dev mailing list