Issue with T.ToUndirected() and heterogeneous data

[maximocrv]

I have followed the guide provided here with a custom dataset, and when I try to call T.ToUndirected()(custom_data) I get the following error: RuntimeError: index -1 is out of bounds for dimension 0 with size 80083, which arises from the call to out.scatter_add_ in torch_scatter/scatter.py. On the other hand T.AddSelfLoops() and T.NormalizeFeatures() seem to work fine, so I am not sure what the issue is. Any ideas as to what the problem could be?

[wsad1]

@maximocrv could you please your data or a sample of data that captures the traits of your data well. That would help debug this faster.

[rusty1s]

This is only possible when installing via pip:

pip install git+https://github.com/pyg-team/pytorch_geometric.git

[shahinghasemi]

Thanks for the replay @rusty1s , I first removed the old one (conda remove pyg) and then executed the above command however now import torch_geometric.nn can't be resolved in my conda environment. What am I missing?
An additional question: What's the reason that you don't update the package semantically? This way it's far convenient to update it using conda update package to use the latest version/commit of it.

[rusty1s]

What do you mean with updating the package semantically? We continuously release new versions with new features and bugfixes. I'll probably release 2.0.3 in the upcoming week.

Can you check that you are indeed installing into the conda directory? What's the output of:

which python
which pip

[shahinghasemi]

What do you mean with updating the package semantically? We continuously release new versions with new features and bugfixes. I'll probably release 2.0.3 in the upcoming week.

I mean updating the package more frequently. for example (if I'm not wrong) RandomLinkSplit or ToUndirected is not updated to the latest commit which are important functions though.

Can you check that you are indeed installing into the conda directory? What's the output of:

which python: /home/shahin/miniconda3/envs/hetero/bin/python
which pip: /home/shahin/miniconda3/envs/hetero/bin/pip

[rusty1s]

I have plans to release 2.0.3 tomorrow. Please stay tuned :)