[
https://issues.apache.org/jira/browse/ARROW-17933?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17655751#comment-17655751
]
Rok Mihevc commented on ARROW-17933:
------------------------------------
# It would be doable to go zero-copy from SparseCOOIndex <->Tensor in 2D, see
[format
here|https://github.com/apache/arrow/blob/3f31b327cd04e79e673b37ee684d438a72367483/format/SparseTensor.fbs#L28-L75].
It's best you're coming from row-major representation. Values are in a
separate buffer and you couldn't really construct a dense table from them
zero-copy.
# We've not had much interest or work in that space. It would probably require
some commitment to get things moving.
> SparseCOOTensor raises error when created with zero elements
> ------------------------------------------------------------
>
> Key: ARROW-17933
> URL: https://issues.apache.org/jira/browse/ARROW-17933
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
> Affects Versions: 9.0.0
> Environment: Ubuntu 22, Python 3.10, Pyarrow 9.0.0
> Reporter: Bruce Martin
> Assignee: Rok Mihevc
> Priority: Major
> Labels: pull-request-available
> Time Spent: 2h 20m
> Remaining Estimate: 0h
>
> Using pyarrow, SparseCOOTensor generates an error when created with zero
> elements. SparseCSCMatrix and SparseCSRMatrix work fine.
> Example:
> {code:java}
> In [24]: pyarrow._version_
> Out[24]: '9.0.0'
> In [25]:
> pyarrow.SparseCSCMatrix.from_scipy(scipy.sparse.csc_matrix(numpy.zeros((2,4)),
> dtype=numpy.float32)).non_zero_length
> Out[25]: 0
> In [26]:
> pyarrow.SparseCSRMatrix.from_scipy(scipy.sparse.csr_matrix(numpy.zeros((2,4)),
> dtype=numpy.float32)).non_zero_length
> Out[26]: 0
> In [27]:
> pyarrow.SparseCOOTensor.from_scipy(scipy.sparse.coo_matrix(numpy.zeros((2,4)),
> dtype=numpy.float32)).non_zero_length
> ---------------------------------------------------------------------------
> ArrowInvalid Traceback (most recent call last)
> Cell In [27], line 1
> ----> 1
> pyarrow.SparseCOOTensor.from_scipy(scipy.sparse.coo_matrix(numpy.zeros((2,4)),
> dtype=numpy.float32)).non_zero_length
> File ~/venv/lib/python3.10/site-packages/pyarrow/tensor.pxi:400, in
> pyarrow.lib.SparseCOOTensor.from_scipy()
> File ~/venv/lib/python3.10/site-packages/pyarrow/error.pxi:100, in
> pyarrow.lib.check_status()
> ArrowInvalid: SparseCOOIndex indices must be contiguous
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
