How can I input two different types of graghs into the network at the same time?

❓ Questions & Help

Hi~ I now have a network input that is two types of graphs，GraphA and GraphB. These two Graphs correspond to the same label. So, How do I input these two types of graphs at the same time? I have two ideas：

1. Put both Graphs in the same Dataloader, But I don’t know how to implement it, it may destroy the graph’s structure inside.
2. Use two Dataloaders for GraphA and GraphB，and do this:

for batch1, batch2 in zip(data_loaderA, data_loaderB):
    batch1, batch2

But, if I set the parameter of dataloader, shuffle=True. These two dataloaders will not generate the same random number, which will cause batch1, batch2 to lose the corresponding relationship, without the same label.

I really need your advice, thank you!

[rusty1s]

I believe the easiest way is to define a new PairDataset:

class PairDataset(torch.utils.data.Dataset):
    def __init__(self, datasetA, datasetB):
        ...

    def __getitem__(self, idx):
        return self.datasetA[idx], self.datasetB[idx]

and use it in a PyTorch DataLoader as follows:

from torch_geometric.data import Batch

def collate(self, data_list):
    batchA = Batch.from_data_list([data[0] for data in data_list])
    batchB = Batch.from_data_list([data[1] for data in data_list])
    return batchA, batchB

torch.utils.DataLoader(pair_dataset, batch_size=..., shuffle=True, collate_fn=collate)

[YunYunY]

Thank you so much. The issue has been solved in discussion.