How can I load 2 different graphs with PyG and pytorch_lighting ?

[xiachenrui]

Hi all,
I am trying to do contrastive learning between two different graphs. Now I bulid two Data objects, but I found that LightningNodeData object can only receive one Data as input. How can I input two graphs at the same time.

Thanks !
Chen-Rui

[xiachenrui]

I need to explain that the nodes of the two graphs are shared, but the edges are completely different, so I cannot use Batch to merge the two graphs

[rusty1s]

In general, DataLoader supports returning tuples of data objects, so something like

class Dataset(...)
    def __getitem__(self, idx):
         return dataset1[idx], dataset2[idx]
         
dataset = Dataset()
loader = DataLoader(dataset, ...)

for batch1, batch2 in loader:
    pass

already works.

You are right that LightningNodeData doesn't support this, but you can consider writing your own DataModule for PyTorch Lightning to support this.

[xiachenrui]

Thanks for your kind reply！