Added sparse concatenation functions

[carlobaldassi]

I noticed that SparseMatrixCSC was lacking support for vcat and hcat and I added that.

[carlobaldassi]

I realized there's a cleaner way to do the sparse hcat (see last commit attached to the pull request).
Also, I should probably implement hvcat as well, I'll look into that.

[StefanKarpinski]

Nice. That is much cleaner.

[carlobaldassi]

Ok I fixed some things:

• generalized to general number of argument (i.e. more than two)
• added support for sparse hvcat
• added a no-op full(x)=x for dense arrays which allows to get the dense cat version as fallback (and which also makes sense imho)

The only drawback w.r.t. the previous version is that concatenation only works with matrices which have the same types (i.e. no more promotion). I'm still trying to find out a workaround for that.

[ViralBShah]

Can you add a couple of tests as well?

[carlobaldassi]

I started writing tests but then I ran into a bug affecting the sparse ref. Here is an example instance:

julia> b = sparse([0.8 0 0; 0 0.3 0.4; 0 0 0])
3x3 sparse matrix with 3 nonzeros:
    [1, 1]  =  0.8
    [2, 2]  =  0.3
    [2, 3]  =  0.4


julia> b[2,3]
0.4

julia> b[2:2,3]
1x1 sparse matrix with 1 nonzeros:
    [1, 1]  =  0.4


julia> b[1:2,3]
2x1 sparse matrix with 0 nonzeros:


julia> #UH?

BTW I solved the promotion issue, it's in the latest commit on the sparse_cat branch in my fork. I'll try to find the bug and then file another pull request with the (hopefully) working tests.

[ViralBShah]

I merged your stuff in, and added some very simple tests. I don't think that sparse ref has been tested at all. I think that with some focus we can get the sparse matrix stuff into good enough shape that it can be documented and people start using it.

Unless you are fixing this right away, do open an issue so that we don't lose it.

-viral

On 15-Mar-2012, at 10:22 AM, Carlo Baldassi wrote:

I started writing tests but then I ran into a bug affecting the sparse ref. Here is an example instance:

julia> b = sparse([0.8 0 0; 0 0.3 0.4; 0 0 0])
3x3 sparse matrix with 3 nonzeros:
  [1, 1]  =  0.8
  [2, 2]  =  0.3
  [2, 3]  =  0.4


julia> b[2,3]
0.4

julia> b[2:2,3]
1x1 sparse matrix with 1 nonzeros:
  [1, 1]  =  0.4


julia> b[1:2,3]
2x1 sparse matrix with 0 nonzeros:


julia> #UH?

BTW I solved the promotion issue, it's in the latest commit on the sparse_cat branch in my fork. I'll try to find the bug and then file another pull request with the (hopefully) working tests.

---

Reply to this email directly or view it on GitHub:
#585 (comment)