Citeseer graph discrepancy

[qubit0]

Why is there a discrepancy between citeseer graph from PyG and DGL?

It seems like PyG gets rid of self-loop from citeseer graph. The extra 124 edges reported by DGL is the number of self-loop in the citeseer graph.

[qubit0]

It seems like PyG did some clean-up. The reason for getting rid of the self-loop is that for the citation network, the self-loop doesn't make sense as it would mean the paper is citing itself, right?

[rusty1s]

Yes, we do some post-processing to clean up the dataset. In the end, it doesn't make a big difference.