User defined minibatching

[Neel-132]

So, I have a set of graphs for link prediction. Now, the way I want to pass a set of graphs in each batch is custom. Also, while passing it to the data loader I want to sample neighbors for a particular set of nodes randomly. More clearly consider this image depicted below:

I want to pass the graph enclosed in the red dotted circle in one batch and while passing I want to sample the nodes associated with the CA1, CA2,... nodes (connected via violet colored edges) randomly. Is there a way to do that without writing custom dataloader?

[wsad1]

Pyg has several dataloaders you can find them here. Check out LinkNeighborLoader and RandomNodeLoader. If none among the loaders fit your use case you might have to write your own dataloader.