tkipf
commented
5 years ago That should work!
On Wed 12. Dec 2018 at 09:00 WowThankyou notifications@github.com wrote:
Hello Thomas,
I have some question about the GCN model. Our training and testing dataset are multiple graph instances, so I build a block-diagonal matrix (adjacency matrix) and concatenate respective feature matrices that you have introduced. However, I want to do node classification task instead of graph classification. In other words, I give labels to each node, and I hope that the GCN model can learn from features in training dataset and then predicts labels to each node in testing dataset.
I am not sure whether the GCN model can do it when there is no edge between training dataset and testing dataset. If it can't could you tell me how to modify or give me some suggestion.
Thank you very much!!
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tkipf/gcn/issues/73, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYG6oP12EzWGAzH2i4C-sw6wSyUJaks5u4LeGgaJpZM4ZO6fD .
WowThankyou
Hello Thomas,
I have some question about the GCN model. Our training and testing dataset are multiple graph instances, so I build a block-diagonal matrix (adjacency matrix) and concatenate respective feature matrices that you have introduced. However, I want to do node classification task instead of graph classification. In other words, I give labels to each node, and I hope that the GCN model can learn from features in training dataset and then predicts labels to each node in testing dataset.
I am not sure whether the GCN model can do it when there is no edge between training dataset and testing dataset. If it can't could you tell me how to modify or give me some suggestion.
Thank you very much!!