gilgamesh: buggy scipy.sparse.linalg

[John Kitchin]

the linalg library performance in python depends a lot on what math
libraries they were linked to and maybe which compiler. the scipy module we
use was built with gcc and versions of blas/atlas/lapack built with gcc.

you may have better luck with performance with a version compiled by ifort
and linked with mkl. scipy is tricky to build, and I went with a build that
worked for most things.

John

-----------------------------------
John Kitchin
Associate Professor
Doherty Hall A207F
Department of Chemical Engineering
Carnegie Mellon University
Pittsburgh, PA 15213
412-268-7803
http://kitchingroup.cheme.cmu.edu




On Mon, May 23, 2011 at 4:41 PM, Asad Hasan <ahasan at andrew.cmu.edu> wrote:

> Hi all,
>
> Has any Python user used scipy.sparse.linalg functions for 'large'
> linear algebra problems, particularly scipy.sparse.linalg.eigen(), on
> gilgamesh ?
>
> It seems that: scipy.sparse linear algebra is much slower than
> comparable MATLAB functions.  Is this expected behaviour ?
>
> I also found that eigenvalues found by scipy.sparse.linalg.eigen() are
> incorrect in some cases (the errors are very large and unrelated to
> iterative convergence).
>
> I am wondering if anyone has come across these problems and how were
> they solved ?
>
> best, asad.
>
> _______________________________________________
> gilgamesh-users mailing list
> gilgamesh-users at lists.andrew.cmu.edu
> https://lists.andrew.cmu.edu/mailman/listinfo/gilgamesh-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.andrew.cmu.edu/mailman/private/gilgamesh-users/attachments/20110523/09fdaec0/attachment.html