Issue with data edge_attr is None

[dgm2]

Hello, Thanks for releasing the code.

I am getting an error in a few datasets (IMDB-B, PROTEINS) about edge_attr being None, as soon as this object is needed in the convolution.
Was this issue encountered before? or any possible suggestions?

Thanks

[beabevi]

Hello giame,

Can you be more specific on the configuration you are running?

[dgm2]

1. I have run data.py with the edge_deletion policy, (also tried some other policies), as follows:
   data.py --dataset=PROTEINS --policies=edge_deleted
   Here it seems that edge_attr is already None, in the dataset, so this seems to cause the issue in the next point.
   This script terminates successfully.

2. in the model, it is required edge_attr: https://github.com/beabevi/ESAN/blob/master/models.py#L186

I run as follows: (model=deepsets, also tried model=gnn)
--dataset=PROTEINS --policy=edge_deleted --model=deepsets --gnn_type=gin
At this point getting error for edge_attr being None

I am running locally with pytorch-geometric 1.7.2

[beabevi]

For the TUDatasets you must use --gnn_type=originalgin.
Indeed, --gnn_type=gin uses edge_attr and it is used for ogbg datasets.

You can see which gnn_type is used for each dataset in the corresponding yaml file under configs/.

[dgm2]

Ah OK I was running with a wrong configuration.
I was able to run with your suggestion.
Thanks a lot!