Matthias Fey

Ok, but this is undefined behavior. If you wanna support `[0, 3, 7]`, then you need to input `x[:, :7]`.

By my comment in https://github.com/pyg-team/pytorch_geometric/pull/9513, I meant why does our CI is still green even without this fix? I am a bit hesitant to merge this without further understanding.

I think this already works as expected. Am I missing something?

`to_data_list()` can currently only be used on unmodified `Batch` objects coming from `from_data_list()`. There is currently no easy way for us to fix this because `subgraph()` is a method at...

We provide `Data.is_node_attr` and `Data.is_edge_attr` functionality which will infer whether attributes are node-level or edge-level vectors. For `Batch.subgraph`: Yes, I think this is possible, but we would require a `mask`...

Yes, you are right, we would also require the `edge_mask`, so we would need to re-structure `Data.subgraph` into: ``` def _subgraph(...): return data, edge_mask def subgraph(...): return self._subgraph(...)[0] ``` Yes,...

For heterogeneous graphs, data leakage is prevent via specifying "reverse" edge types: ```python edge_types=("user", "rates", "movie"), rev_edge_types=("movie", "rev_rates", "user") ``` This makes sure that links are eliminated in the reverse...

You need to install PyTorch from the official source, not from `conda-forge`.

It works for me. Can you do me a favor and modify [this line](https://github.com/pyg-team/pytorch_geometric/blob/2.5.3/torch_geometric/nn/conv/message_passing.py#L196) to ```python except Exception as e: print(e) ``` and see what comes back?

It looks like downloading the files failed in your case. Can you remove the dataset `root_dir` and try again? It works fine for me.