Can Graphormer be modified to accommodate node classification and link prediction problems?

[zsz00]

Can Graphormer be modified to accommodate node classification and link prediction problems?

Can you give a few examples?

[Austinzhenghua]

is this question solved?

[LOUGBERTI]

可以修改 Graphormer 以适应节点分类和链接预测问题吗？

你能举几个例子吗？ 你好，可以用到链接预测问题吗？

[mavisguan]

Hi, sorry for the late response. For the node classification problem, you can modify Graphormer's forward method to produce predictions for each node. The GraphormerGraphEncoder's forward method will return graph_node_feature (contained in the variable x) in inner_states, and you can use it to predict node features. For now Graphormer can't be used to solve the link prediction problem directly, but the variable attn_bias might be helpful if you want to develop a Graphormer variant yourself to solve the problem. You can use the debugging tool I mentioned in issue #133 to see each tensor's shape and guess each tensor's meaning. We might provide some example code in the future, thanks for your suggestions.