[SciPy-User] "Zero"-shape sparse matrices

Jaakko Luttinen jaakko.luttinen@aalto...
Tue Feb 21 09:58:37 CST 2012


On 02/21/2012 05:00 PM, Warren Weckesser wrote:
> 
> 
> On Tue, Feb 21, 2012 at 7:23 AM, Jaakko Luttinen
> <jaakko.luttinen@aalto.fi <mailto:jaakko.luttinen@aalto.fi>> wrote:
> 
>     On 02/17/2012 06:48 PM, Christopher Mutel wrote:
>     > On Fri, Feb 17, 2012 at 4:32 PM, eat <e.antero.tammi@gmail.com
>     <mailto:e.antero.tammi@gmail.com>> wrote:
>     >> On Thu, Feb 16, 2012 at 5:44 PM, Jaakko Luttinen
>     <jaakko.luttinen@aalto.fi <mailto:jaakko.luttinen@aalto.fi>>
>     >> wrote:
>     >>> To make a long story short, Scipy doesn't seem to allow sparse
>     matrices
>     >>> that have length zero on any of the axes. For instance:
>     >>> C = numpy.ones((0,0))
>     >>> K = scipy.sparse.csc_matrix(C)
>     >>> ValueError: invalid shape
>     >>>
>     >>> It is possible to create a "zero"-shape dense matrix but not sparse.
>     >>> Why? To me, this seems like a bug.. Is it so?
>     >
>     > I am not an expert, but it is my understanding that the sparse matrix
>     > implementations in SciPy assume precisely two dimensions. One
>     > dimension having a size of 0 would break all the assumptions of this
>     > code. The NumPy array class is a much more generic container, and was
>     > designed from the beginning to allow a number of slicing and
>     > dimensionality tricks (see the documentation on numpy striding). You
>     > can search through the mailing list from a few years ago to find a
>     > discussion about three dimensional sparse matrices, and the conclusion
>     > was the same: SciPy supports 2-d (in the sense of two real dimensions)
>     > sparse matrices only.
> 
>     Thanks for your answer!
> 
>     I think a matrix with shape (10,0) would be as "2-d" as a (10,1) shaped
>     matrix. Both have two dimensions, but neither one has both axes longer
>     than 1. I don't mean to consider 0-d, 1-d, 3-d or N-d matrices, but
>     empty 2-d matrices. I just don't see why there is this limitation for
>     sparse matrices that zero is not a valid length for an axis.
> 
> 
> 
> In principle, sparse matrices should behave as much like dense matrices
> as possible.  Since numpy ndarrays and matrices allow a dimension to
> have length 0, it is a reasonable expectation for sparse matrices to
> allow this also.  Could you file a ticket for this? Click on the "bug
> reports" link here: http://projects.scipy.org/scipy

Ok, thanks!
http://projects.scipy.org/scipy/ticket/1602

-Jaakko


More information about the SciPy-User mailing list