[SciPy-user] Fastest way to save/load fairly large (sparse) matrices

[David Warde-Farley]

Hi folks,

I've been wrestling with this problem for some time now. Essentially  
I'm using Numpy/Scipy as the backend in a python-based web  
application and need a way of serializing the contents of a fairly  
large, but sparsely populated, symmetric matrix to disk. So far I've  
experimented with just ASCII dumps and pickling, but neither really  
has the kind of speed that I'm looking for (and is kind of necessary  
for an interactive web application). Is there some better way?

To give you an idea of the scale, I'm hoping to get this to work with  
20000x20000 matrices that are about 1% full.

Thanks,

David