[C++] Sparse tensor naming

[asfimport]

Currently there's SparseCOOIndex and SparseCSRIndex, but also SparseTensorCOO and SparseTensorCSR.

For consistency, it would be nice to rename the latter SparseCOOTensor and SparseCSRTensor.

Also, it's not obvious the SparseMatrixCSR alias is useful.

Reporter: Antoine Pitrou / @pitrou
Assignee: Kenta Murata / @mrkn
Watchers: Rok Mihevc / @rok

PRs and other links:

• GitHub Pull Request #5605

_{Note: This issue was originally created as ARROW-6671. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
cc @mrkn

[asfimport]

Kenta Murata / @mrkn:
Indeed. I want to make their name consistent, so I’ll make a pull request tomorrow.

@pitrou How about employ SparseCSRMatrix rather than SparseCSRTensor because it cannot represent a tensor with more than 2-dimension.

[asfimport]

Antoine Pitrou / @pitrou:
I don't know. We don't use the term "matrix" currently in Arrow. cc @wesm

[asfimport]

Rok Mihevc / @rok:
SparseCSRMatrix might be more misleading as it 'doesn't look look like' a Tensor type. I think that is potentially more confusing than it being limited to 2D.

+1 for the consistent naming

[asfimport]

Wes McKinney / @wesm:
I am OK with having the type be called "matrix". Good to make things as clear and consistent as possible

[asfimport]

Kenta Murata / @mrkn:
There are three tickets about adding new sparse tensor index types:

• ARROW-4225: Add CSC sparse matrix support

• ARROW-4226: Add CSF sparse tensor support

• ARROW-6485: Support the format of a COO sparse matrix that has separated row and column indices

  In ARROW-6485, we will introduce another COO sparse index that manages indices in two vectors, the same format as scipy.sparse.coo_matrix, rather than a single matrix, that is the current SparseCOOTensor's format.

  If we can use both "matrix" and "tensor" for sparse index type names, we will have the following sparse index types after the above tickets are closed:

• SparseCOOTensor

• SparseCSFTensor

• SparseCOOMatrix

• SparseCSRMatrix

• SparseCSCMatrix

  By looking at the type names, we can recognize what types can represent tensors with more than 2 dimensions.

[asfimport]

Antoine Pitrou / @pitrou:
Issue resolved by pull request 5605
#5605