pratyushadk opened a new pull request, #50907:
URL: https://github.com/apache/arrow/pull/50907

   ### Rationale for this change
   
   Calling `SparseCSRMatrix::Make` or `SparseCSCMatrix::Make` on a 1D dense 
tensor currently returns a `NotImplemented` error ("TODO for ndim <= 1"). The 
CSR/CSC representation for a 1D vector is well-defined though: it is equivalent 
to a single-row sparse matrix where `indptr = [0, nnz]`, `indices` holds the 
positions of non-zero values, and `values` holds the corresponding values. The 
COO format already handles 1D tensors, so this fills a straightforward gap.
   
   ### What changes are included in this PR?
   
   `cpp/src/arrow/tensor/csx_converter.cc`
   
   The `Convert()` method in `SparseCSXMatrixConverter` now handles three cases 
cleanly. `ndim == 0` returns `Status::Invalid` since scalars have no meaningful 
sparse representation. `ndim == 1` takes a new path that does a single linear 
scan of the vector, writes `indptr = [0, nnz]`, and records the positions and 
values of non-zeros. `ndim == 2` is the existing path and is completely 
unchanged.
   
   The reverse path (`MakeTensorFromSparseCSXMatrix`) had a hardcoded 
`shape[1]` access which would be out of bounds for a 1D tensor. That is fixed 
to use `shape.size() > 1 ? shape[1] : shape[0]`.
   
   `cpp/src/arrow/sparse_tensor_test.cc`
   
   Three new tests added to the existing `TestSparseCSRMatrix` fixture covering 
the 1D case.
   
   ### Are these changes tested?
   
   Yes. Three new tests cover the main conversion path, the all-zero edge case, 
and a full round-trip from dense to sparse and back. All 153 tests in the 
sparse tensor test suite passed locally with no regressions.
   
   ### Are there any user-facing changes?
   
   `SparseCSRMatrix::Make` and `SparseCSCMatrix::Make` now accept 1D tensors 
instead of returning `NotImplemented`. There are no breaking changes.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to