[Numpy-discussion] np.dot and scipy sparse matrices

Stéfan van der Walt stefan@sun.ac...
Thu Nov 10 12:56:39 CST 2011


On Thu, Nov 10, 2011 at 10:45 AM, Peter Prettenhofer
<peter.prettenhofer@gmail.com> wrote:
>> I don't know if np.dot is supposed to know anything about sparse
>> matrices; if so, then a fancier change may be needed.
>
> Ok, then I'll pass that on to the (sparse) scipy folks.
> Are there any alternative to np.dot you are aware of?

With sparse matrices, you can simply do

S = ss.csr_matrix(np.arange(9).reshape((3,3)))
x = np.array([[1],[1],[1]])
print S * x

[[  6.]
 [ 15.]
 [ 24.]]

But I assume what you want is an easy way to handle both sparse
matrices and dense arrays with the same call?  I noticed in your
comment on the sklearn list that this used to work on older versions
of numpy.  If that is the case, this is probably a regression that
needs to be fixed.

Regards
Stéfan


More information about the NumPy-Discussion mailing list