light_gcn graph micro batching

[sorushsaghari]

Hi everyone. torch_geometrics has a lightgcn model implemented which on the forward method has an edge_index param.
Because my dataset is large I can't load it into memory. is there a solution like loader.DataLoader to micro-batch edge_index. or this neural network need whole data at the same time and I can't do anything

[akihironitta]

Hi @sorushsaghari, you can utilise NeighborLoader (or LinkNeighborLoader) for your use-case. Have you had a look at the docs by any chance? https://pytorch-geometric.readthedocs.io/en/2.4.0/tutorial/neighbor_loader.html

[sorushsaghari]

it seems to be a good solution. so you mean i can sample edge_index for each node i want to train ? and reduce my edge_index size?

[akihironitta]

Yes.