Hello!

Most of them are easy to understand whether they are based on synthesis or loop graph neural networks, or based on spectral domain and spatial domain. They are explained by message passing or spectral graph theory, but those based on LSTM (GRU) are not very well understood. I would like to ask you to explain its meaning or principle. Thank you very much.

GNN_Implement_with_Pytorch -> Linear GNN模型 -> line175 :

H = torch.index_select(H, 0, X_Node)

it should be:

H = torch.index_select(H, 0, X_Neis)

when i run :
train_loss, train_acc, test_acc = train(node_list=list(map(lambda x:x[0], N)),
edge_list=E,
label_list=list(map(lambda x:x[1], N)),
T=5)
the error shows like below:
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [2, 2]], which is output 0 of TBackward, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

pytotch 1.8.0