A question about the EdgeDataLoader

Questions
, ,
#1

Hi, I have a question about the EdgeDataLoader https://github.com/dmlc/dgl/blob/master/examples/pytorch/graphsage/train_sampling_unsupervised.py#L216
For each edge with ID e in Reddit dataset, the reverse edge is e ± |E|/2
I tried that in reddit data:

n_edges // 2
57424428

so edge_id0 should have the reverse edge with edge_id57424428, but I found that the node pairs are reversed:

g.find_edges(0)
(tensor([0]), tensor([225202]))

g.find_edges(57424428)
(tensor([100148]), tensor([245]))

Did I make a mistake?

#2

That’s a bug. The self_loop argument in RedditDataset should be False. Good catch.