[SciPy-dev] Possibly bug in sparse.py

[Robert Cimrman]

Nils Wagner wrote:
>     C =dot(A,XX)+dot(XX,B)
>   File "/usr/local/lib/python2.4/site-packages/scipy/sparse/sparse.py",
> line 532, in __add__
>     raise ValueError, "inconsistent shapes"
> ValueError: inconsistent shapes
> 
>>>>A
> 
> <317x317 sparse matrix of type '<type 'float64scalar'>'
>         with 7327 stored elements (space for 7327)
>         in COOrdinate format>
> 
>>>>XX
> 
> array([[0, 1, 1, ..., 1, 1, 1],
>        [1, 0, 1, ..., 1, 1, 1],
>        [1, 1, 0, ..., 1, 1, 1],
>        ...,
>        [1, 1, 1, ..., 1, 1, 1],
>        [1, 1, 1, ..., 1, 1, 1],
>        [1, 1, 1, ..., 1, 1, 1]])
> 
>>>>shape(XX)
> 
> (317, 10)
> 
>>>>shape(B)
> 
> (10, 10)
> 
> 
>>>>B
> 
> <10x10 sparse matrix of type '<type 'float64scalar'>'
>         with 28 stored elements (space for 28)
>         in Compressed Sparse Column format>
>  
> How can I fix this problem ?

First, use A * XX + XX * B instead of dot, since dot is a numpy.core 
function and as such cannot handle sparse matrices.

Then there was a bug in the coo_matrix() constructor (not fixed yet in 
svn, cannot log in...) causing that the shape was one less than the 
actual shape... Thanks for debugging!

---
Side note, not directly related to your problem:

Another bug in the coo_matrix() constructor is that it does not work for 
all value types (e.g. int32), because some dtypes are probably missing 
in the type translation table of sparse.py - can someone (Travis? :) 
help with this? I have followed the recent changes to the dtype object 
only superficially...

r.