[SciPy-user] coo_matrix, lapack, cvxopt

[Nathan Bell]

On Mon, Mar 23, 2009 at 8:59 PM, Eric Friedman <ejf27 at cornell.edu> wrote:
> Hi, a couple of questions:
>
> 1) how do I look inside a coo_matrix -- I'd like to extract the i,j, data?

If A is your coo_matrix, you can grab the COO arrays using:
I = A.rows
J = A.cols
V = A.data

>
> 2) I need to solve large sparse rectangular matrix equation which doesn't have
> full rank, but is solvable.  spsolve (dsolve) doesn't work.  Can I do my own LUP
> decomposition using scipy.sparse?

I believe the UMFPACK scikit supports that functionality.

> 3) How does lapack compare to umfpack? (I'm thinking about using cvxopt which
> has lapack.)  Also more generally, how does cvxopt's sparse routines do compared
> to scipy?

Unknown.  There are not so many sparse factorization methods (SuperLU,
UMFPACK, Taucs), so it's likely that you'll get the one of them.

-- 
Nathan Bell wnbell at gmail.com
http://graphics.cs.uiuc.edu/~wnbell/