[SciPy-user] Sparse Compilation Issues (again)

[Travis Oliphant]

>
> Nope, straight checkout from CVS.  I had the same thought as you, so I
> checked it out twice with the same results.  I also looked for SuperLU
> itself, but the only version I found out there was 2.0 which (surprise)
> doesnt compile with sparse (I think it's bundled with 1.2?)...
>

I remember upgrading SuperLU to version 2.0 just a couple of months ago.
Is the CVS version not 2.0?  Let me check into that.

> Another question that's been bugging me now that I've been staring at
> sparse's source code... it looks to me like it's just a sparse matrix
> equation solver.  I will need to instantiate very large "cubes", say
> 1e3x1e3x1e4, with low population densities (max around 20%, typically
> 5-10%).  This 1e10 cube doesn't even allocate in Numpy/Scipy (smaller cubes
> actually crash my Win2K laptop), so I was _hoping_ that scipy.sparse would
> implement a "default value" array: i.e. the operation:
> x = zeros([1e3,1e3,1e3])

Ahh.  You want something else then.  Sparse is not a sparse array, it is
only a sparse matrix.   Sparse matrices have been well studied and there
are many good storage mechanisms and algorithms using them.

Sparse arrays, on the other hand are usually just implemented as
dictionaries.   I know of no sparse array implementation for any
array-oriented environment.  You could make your own without too much
difficulty, supporting just the operations you care about.

Which operations do you want to do on these "sparse arrays"?

Really, the sparse matrix module in SciPy does not give you much if what
you really want is sparse arrays.

I'd love to hear of any successes you have in this.

-Travis