Divide a single graph into batches

[mhadnanali]

Can we divide a single graph into batches any working example? E.G DBLP dataset I can not execute with 24 GB GPU {with GRACE, Deep Graph Contrastive Representation Learning }

It is mentioned "PyG automatically takes care of batching multiple graphs into a single giant graph with the help of the torch_geometric.loader.DataLoader class."

but what about one graph into mini-batches? is it feasible or not? if not then why (for my knowledge)

[EdisonLeeeee]

It is feasible to divide a single graph into batches. The simplest way is by NeighborLoader, which allows mini-batch training on large-scale graphs where full-batch training is not feasible. Also, I think ClusterLoader is what you want to get different batches from a single graph.

[jiaqian]

I am wondering if batching step destroy a connected graph, say G, into several batches? It will make some difference as running the whole graph G on one BIG batch? Does anyone have any idea?

[hahahaha90]

Hi, could you share the dblp dataset? thanks!

[rusty1s]

It is available at torch_geometric.datasets.DBLP.

[mhadnanali]

visit this page, you are going to find all of the datasets here. https://pytorch-geometric.readthedocs.io/en/latest/modules/datasets.html