A question about GCN

[jjjj0x]

Hi tkipf ,

I try to use GCN on my own data to classify the nodes into category one [0,0] and category two[0,1]. And what happen is that the model classify all the nodes into one category even after I greatly reduce this category's ratio in train set. I can't figure out why, so I make a simpler data set (a graph), which including six nodes, where the node 1 and 2 , 3 are fully connected, while, node 4,5 and 6 are connected. The 1 is labeled in + and 4 is labeled in -. The test set is including of 2 and 6. After I train GCN on this data, the outputs is just like I thought. All the nodes are classified into one category even 1,2,3 and 4,5,6 are not connected. Please, give me some advices on this. I don't understand how this happened.

[qiubin1997]

你好，请教一下你是怎么把GCN模型在自己的数据集上运行的，是使用的tkipf作者的GCN模型吗？我调试了很久，都没能在自己的数据集上运行。希望你可以指点一下，我的研究方向是交通流预测，希望我们可以探讨。期待您的早日回信，万分感谢! 

我的邮箱[email protected] ，qq也是这个。

 祝您一切安好