[SciPy-dev] small benchmark of LP solvers [from GSoC project]

[Joachim Dahl]

On 6/4/07, dmitrey <openopt at ukr.net> wrote:
>
>  Yes, of course, I had noticed the cvxopt feature.
> So I decided to transform matrix to cvxopt sparse matrix if
> nnz(A)/numel(A)<0.3, as I had seen the recommendation somewhere in matlab
> sparse stuff
> (and I wonder why cvxopt developers hadn't do something like that by
> themselves (like glpk and lp_solve do), it consumes 2 lines of code in my
> CVXOPT_LP_Solver.py:
>

Not all optimization problems are sparse.  In particular many engineering
problems are dense,  in which case you want to use dense BLAS/LAPACK.

You can just download a 30 day trial version of MOSEK.  It's quite easy,
and
their solvers are terrific at exploiting sparsity, it exploits
multi-processors,  and
the next version will have a native Python interface.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20070604/a035fd12/attachment.html>