How to run explainers on Planetoid datasets?

[smtnkc]

Thanks for your efforts on this work!

I understood the examples you provided for the MUTAG TUDataset. I need to run explainers for Planetoid datasets. More specifically, I want to implement test_SubgraphX_Cora.py, which is very similar to test_SubgraphX_MUTAG.py.

Can you provide a guideline for this purpose?

[owencqueen]

Hello,

We're glad that you're interested in using our codebase! If you are simply substituting another PyTorch Geometric dataset, it should work in a similar way to the script you cited here in our repo, just replace the following line (line 16 from the test_SubGraphX_MUTAG.py script):
dataset = TUDataset(root='data/TUDataset', name='MUTAG')
with a line to load your dataset. Hopefully that helps, but please provide more details if you need more assistance beyond this, we're happy to help!

Thanks,

GraphXAI team

[smtnkc]

@owencqueen Thanks for your reply.

Unfortunately, just changing the dataset loaded does not work in this case because the structure of the TUDataset and Planetoid is different.

I deep-dived into the codebase, and what I see is that test_SubgraphX_MUTAG.py uses MUTAG.py to create a MUTAG dataset object which is a GraphDataset. However, Planetoid datasets are NodeDatasets (which consist of a single graph rather than multiple).

I have tried to create a CORA.py similar to MUTAG.py and modify the test script accordingly. However, I could not manage it since the test script is very specific to MUTAG (or GraphDataset in general).

So, my question actually is that how can I run the explainers for NodeDatasets?

Thanks :)