[Sparse] Raise exception when expand is called on sparse tensor

[malfet]

It's already not working, but this makes error message a bit more readable. I.e. it turns:

% python -c "import torch;x=torch.eye(3).to_sparse().expand(3,3)" 

from

NotImplementedError: Could not run 'aten::as_strided' with arguments from the 'SparseCPU' backend.

to

RuntimeError: Expand is unsupported for Sparse tensors.

cc @alexsamardzic @nikitaved @pearu @cpuhrsch @amjames @bhosmer

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/98365

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Windows CPU runner long queue

✅ No Failures

As of commit ef8049c:
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[malfet]

@cpuhrsch do you know why why torch.eye(3).to_sparse_csr().is_sparse() returns false?
Also, why python -c "import torch;print(torch.eye(3).to_sparse().stride())" prints (0, 0), but the same command for sparse_csr tensor raises runtime exception?

[pearu]

do you know why why torch.eye(3).to_sparse_csr().is_sparse() returns false?

@malfet This is because .is_sparse attribute is for testing COO sparse tensor, not for CSR sparse tensor:

>>> torch.eye(3).to_sparse_csr().is_sparse
False
>>> torch.eye(3).to_sparse_csr().is_sparse_csr
True

(.is_sparse is not called .is_sparse_coo because of BC)

Also, why python -c "import torch;print(torch.eye(3).to_sparse().stride())" prints (0, 0), but the same command for sparse_csr tensor raises runtime exception?

sparse compressed tensor implementation defines sym_strides_custom (that raises the runtime exception) while the sparse coo tensor implementation doesn't. IIRC, there was an edge case where coo stride() result was actually used but I couldn't find it quickly (enabling the exception for coo tensor strides might give a hint). In any case, sparse tensors don't have strides and one should not call the method on sparse tensors in general.

[malfet]

do you know why why torch.eye(3).to_sparse_csr().is_sparse() returns false?
(.is_sparse is not called .is_sparse_coo because of BC)

@pearu then perhaps the documentation should be updated, because according to it: Tensor.is_sparse "is True if the Tensor uses sparse storage layout, False otherwise."

sparse compressed tensor implementation defines sym_strides_custom (that raises the runtime exception) while the sparse coo tensor implementation doesn't. IIRC, there was an edge case where coo stride() result was actually used but I couldn't find it quickly (enabling the exception for coo tensor strides might give a hint). In any case, sparse tensors don't have strides and one should not call the method on sparse tensors in general.

Let me dry that in separate PR, as I believe stride() results indeed does not make sense for any type of sparse tensors.

[amjames]

do you know why why torch.eye(3).to_sparse_csr().is_sparse() returns false?
(.is_sparse is not called .is_sparse_coo because of BC)

@pearu then perhaps the documentation should be updated, because according to it: Tensor.is_sparse "is True if the Tensor uses sparse storage layout, False otherwise."

IMO: We should break that compatibility. We are still operating under the beta tag, so we can do that. We really just need to strict about BC with changes that affect other APIs (like gradcheck)

The docs also probably need a review to make sure that any place where "sparse layout" refers specifically to "sparse_coo" is clarified.

[pearu]

LGTM! Thanks, @malfet!

[pearu]

Yes, the docs need an update.

In fact, I think we should just deprecate the usage of .is_sparse and .is_sparse_csr attributes (notice that there are no .is_sparse_csc/bsr/bsc attributes) in favor of using the .layout attribute.

That is, instead of x.is_sparse, use x.layout is torch.sparse_coo, etc.

For example, we often use x.layout in {torch.sparse_coo, torch.sparse_csr, torch.sparse_csc, torch.sparse_bsr, torch.sparse_bsc} to test if x has any of the sparse layouts.

[malfet]

@pytorchbot merge

[pytorchmergebot]

Merge failed

Reason: Approval needed from one of the following:
dhthompson, soumith, jisaacso, osalpekar, asalioufb, ...

Details for Dev Infra team

Raised by workflow job

Failing merge rule: Core Maintainers

[malfet]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

[pytorchmergebot]

The merge job was canceled. If you believe this is a mistake,then you can re trigger it through pytorch-bot.

[malfet]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here