Error encountered in NeighborLoader for own dataset

[xubingze]

Hi there! I am trying to do node classification for my own dataset. There are 7000+ nodes in the network, but only 1000 of them contains label. So I define mask for the nodes that have labels.

train_loader = NeighborLoader(data_stock, input_nodes=torch.tensor(data_stock.train_mask, dtype=torch.bool),
                              num_neighbors=[3,2], batch_size=256,
                              directed=False, shuffle=True)
print(data_stock)
print(data_stock.train_mask)
print(torch.tensor(data_stock.train_mask, dtype=torch.bool))
for batch in train_loader:
    print(batch)

However, I encounter this error and I have no ideas why this happen since I have converted the type of data_stock.train_mask. Thank you very much for your help!

[wsad1]

Convert all your masks in data_stock like train_maskto torch tensors. That should fix this.

[xubingze]

Thank you very much for your reply! I tried your suggestion, but the error still occurred. I plan to use NeighborLoader only in training not in testing process, so I am not sure why the val_mask or test_mask can lead to this error message. Here is the screenshot:

[wsad1]

Are there any other lists in data_stock. Whats the type of x, edge_index and y?

[xubingze]

Yes, I checked the type of x, edge_index and y. The first two are tensor type, but the y are generated from a list (each node is classified as a number between 0 and 5). After I converted y as the type of torch.int64, the error is resolved. Thank you very much!