ENH: allow addition/substraction of 0 to a sparse matrix

[dlax]

No description provided.

[jakevdp]

I'd change the error text to specify "adding a nonzero scalar..."

[jakevdp]

Also, this behavior may lead to some confusion: with dense matrices, adding zero returns a copy. Sparse matrices should probably behave the same way to be consistent.

[dlax]

Jacob Vanderplas wrote:

I'd change the error text to specify "adding a nonzero scalar..."

Also, this behavior may lead to some confusion: with dense matrices, adding zero returns a copy. Sparse matrices should probably behave the same way to be consistent.

Done.

[jakevdp]

Out of curiosity, what use case do you have in mind for this? I can't think of any case where I'd want to add a scalar to a matrix, raising an error unless the scalar is zero. That would essentially require me to explicitly check to see if the scalar is zero before the operation. But if I know it's zero, why would I do this operation, except as a roundabout path to a matrix copy?

[dlax]

Jacob Vanderplas wrote:

Out of curiosity, what use case do you have in mind for this?

sum(a_list_of_sparse_matrices)

[jakevdp]

Makes sense!

[dlax]

I'd like to merge this soon unless somebody objects.

[jakevdp]

Sorry - I lost track of this in all the buzz around pycon!
Before merge, we'll need a unit test for this feature.

[dlax]

Jacob Vanderplas wrote:

Before merge, we'll need a unit test for this feature.

Test added.

By the way, there's a test in test_construct/test_diags_bad that
appears to fail in this branch but not in master. I'm not sure why this
happens but this may be related the merge of #167.

[jakevdp]

Nice work on the test, I'll merge this now.
If you could write the scipy-dev list with the traceback for the test_diags_bad error, that would be very helpful.

[dlax]

Jacob Vanderplas wrote:

If you could write the scipy-dev list with the traceback for the test_diags_bad error, that would be very helpful.

The test failure was only in the topic branch. Now that you've merged
into master, there's no more failure.