Get global edge indexes after getting mini batches through the Dataloader

[purvanshi]

I want to map the edge indexes from every batch to the original edge indexes. Do you know if there is any internal mapping dictionary present that we can access?

[EdisonLeeeee]

There is a e_id attribute in each batch, which maps each edge to the original edge indexes.

[rusty1s]

(important to note that this was just recently added). For PyG <= 2.2, you can do

data.e_id = torch.arange(data.num_nodes)
loader = NeighborLoader(data, ...)