can cvxopt.lapack.gels handle sparse matrices?

[dmitrey]

hi CVXOPT developers,
can cvxopt.lapack.gels handle sparse cvxopt matrices?
Does lapack.gels take benefit of sparsity?

Thank you in advance,
D.

[Joachim Dahl]

no, the cvxopt.blas and cvxopt.lapack modules are just wrappers of BLAS
and LAPACK,
which are for dense matrices.

We implemented a few BLAS-like functions in cvxopt.base, but we did not
intend it to be
a complete sparse BLAS alternative for the users; we only implemented a
few functions we
frequently need in the convexs solvers.

dmitrey skrev:

[dmitrey]

Ok, I understood.
Since there are no sparse LLSP solvers in OpenOpt for now, I intend to
create a converter that will translate sparse LLSP problems to sparse
QP problems and then invoke cvxopt qp solver.
However, could you inform haven't you willing in future to add native
LLSP solver into CVXOPT0? - This is convex problem, isn't it?
Regards, D.

[Joachim Dahl]

If it is a traditional least-squares problem without constraints,
couldn't you solve it directly using either the sparse QR or Cholesky
factorization in SuiteSparse?

We do not currently include the QR factorization in SuiteSparse,
but we have talked about including it in a future release.

Joachim

[dmitrey]

I meant general lineary constrained linear least squares problems, or,
at least, box-bounded ones.
I had tried to connect toms587 (R. J. HANSON and K. H. HASKELL) but a
problem with f2py had been encountered
http://groups.google.com/group/scipy-user/browse_thread/thread/bb3ad277e9213d3e#
And I don't know any other Python code for constraied LLSP for now.

On 10 Чер, 13:53, Joachim Dahl <dahl.joac...@gmail.com> wrote:
> If it is a traditional least-squares problem without constraints,
> couldn't you solve it directly using either the sparse QR or Cholesky
> factorization in SuiteSparse?

Mb I could try, but I guess it would be done better by more
experienced in CVXOPT people.

Regards, D.