Matthias Fey

Not yet, sorry :(

Looks like you installed `torch-cluster` without CUDA support. How did you install it in the first-place?

If you look at the installation path of `torch-cluster` in `site-packages/`, do you see the binaries for CUDA? This should be called something like `*_cuda`.

Can you try to manually remove the `@torch.jit.script` decorators in `torch_cluster/radius.py` in your installed version and report whether this resolves the issue?

We currently have trouble building conda wheels on Windows, that's why it is probably downgrading you to an earlier PyTorch version. Would installing from `pip` work for you for now?

Yes :(

Yes, I am sorry. We don‘t have any MPS-specific implementation right now. You might be able to convert to CPU before calling the cluster routines, and back to MPS afterwards.

Thanks for reporting. Indeed, currently `graclus_cluster` assumes a well-defined graph. We could probably error out early in this we see empty inputs.

I identified the issue in the SAGPool `layers.py` code: ```python score = self.score_layer(x, edge_index).squeeze() ``` will actually squeeze the tensor to a 0-dimensional tensor if the input has shape `[1,...

I'll take a look.