How to organize pairs of subgraphs in social networks into batch data for training?

[finquick]

In the social network, I regard each node as a person and extract the first-order neighbor into a subgraph, the source network and the target network are subgraph extraction, and then compare the similarity of the two subgraphs, the current practice is not organized into the batch, can not use PYG accelerated training characteristics, is there a reference to imitate the demo, thank you!

[wsad1]

For pairs of graphs you need to modify the __inc__ in dataloader as described here