Also, since we don't have a hard dependency on scipy, maybe it would be better to make another function?

Sometimes it is better to use sparse and sometimes it is better to use
dense. We could try to include code that chooses, but perhaps its better
to have the user decide.

Also, in looking at the PR, there is a function smat used. What is that?
or did I read it wrong.
Dan

On Sun, Oct 6, 2013 at 8:35 PM, chebee7i notifications@github.com wrote:

Also, since we don't have a hard dependency on scipy, maybe it would be
better to make another function?

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/969#issuecomment-25780390
.

@chebee7i @dschult Apologies, i just missed copying over the smat() from my implementation

• Have made the use of sparse matrix optional.
• Do we still need to separate this functionality?

Can we just use to_scipy_sparse_matrix() instead of smat?

There is discussion in #913 about using sparse matrices for all of the linear algebra operations in NetworkX.

My first attempt was to utilize to_scipy_sparse_matrix() but it resulted in an exception which I could not debug and had to proceed with my own implementation. I can attempt to recreate that exception if necessary.

Using sparse matrices would indeed be a good idea. This is an implementation that I have been using for my own work which makes it applicable for large graphs. So I thought I'd share it.

Using the sparse eigensolver is probably the right approach for large enough problems and I like the approach.
Could you post the exception so we can figure out how to fix the to_scipy_sparse_matrix() function?

@hagberg Issue opened w.r.t to_scipy_sparse_matrix()

thanks

Can this be updated to use to_scipy_sparse_matrix() ?

Yeah, I'm on it.