SparseMatrixCSR should be a type alias for Transpose{SparseMatrixCSC}

[ndinsmore]

I am surprised that this isn't already in SparseArray, but should we define:
const SparseMatrixCSR{Tv,Ti} = Transpose{Tv,SparseMatrix{Tv,Ti}} where {Tv, Ti <: Integer}

This plus a few simple additional methods would make it so Julia had psuedo built in support for compressed sparse matrixes.

I am already using a definition like this and I can't believe that I am the only one.

If people agree I can take a shot at the PR.

[KlausC]

Do you mean Transpose or Adjoint? Note, that adjoint(transpose(A)) == conj.(A) for complex element type. What would be the benefit of having this constant? There is currently not much support for operations on Transpose{ SparseMatrixCSC}`.

[ndinsmore]

I was originally using Adjoint but I don't think that it is semantically correct. Since this is basically a data manipulation operation rather than a mathematical one, which I think points to Transpose particularly for data types like complex.

This would necessitate some built out of operations on Transpose{ SparseMatrixCSC} but I would like to think that shouldn't be too excessive.

The benefit would be that it would standardize the ecosystem to using the SparseArrays when the have a problem that necessitates a CSR matrix, rather than an ad-hoc solution. This will also ensure that in those cases users that need CSR will have a solution that this performant out of the box.

[KlausC]

This would necessitate some built out of operations on Transpose{ SparseMatrixCSC} but I would like to think that shouldn't be too excessive.

It would, believe me. I am working on general wrappers of sparse matrices for a while now. Best what was recently merged into julia1.2 are conversions of wrapped sparse to SparseMatrixCSC in order to be able to call the specialized methods of this type.

[ViralBShah]

I am inclined to close this one, as a substantial rewrite that results out of this should perhaps be done in an external package.

[ndinsmore]

I think that at it's simplest this would only be a single lined alias, with a few more lines to for a "constructor function" and convert. Give me a few weeks to put together the basic PR

[ViralBShah]

If we introduce this, pretty much everything has to work on it and we have to field issues around performance that come up. I would be curious to see how far we can get.

[martinholters]

Is the recursive nature of transpose an obstacle here?

[ViralBShah]

I think this can easily be experimented with in a package called SparseMatrixCSR.jl.