Baselines

[DongXiang-CV]

Hi authors,

Could you please provide the implementations of deeper baselines such as JKNet (64) and Incep (64) ?

Thank you very much!

[chennnM]

I used the code from DropEdge, including Incep and JKNet.

[DongXiang-CV]

Thanks a lot. I have another question about the Figure 1. Semi-supervised node classification accuracy v.s. degree.

In the main body of the paper, you group the nodes of each graph according to their degrees. The i-th group consists of nodes with degrees in the range [2^i, 2^{i+1}). The x-axis 10^{1}, 10^{2}, 10^{3} in your figure denote index $i$ right?

[chennnM]

The x-axis 10^{1}, 10^{2}, 10^{3} denote degrees. We don't show the value of $i$ in the figure. The eight points on each line are 2^{1}, 2^{2}, ..., 2^{8}.

[DongXiang-CV]

Thanks for your reply.

[DongXiang-CV]

I have some questions on Figure 1. Semi-supervised node classification accuracy v.s. degree.

Do you plot the accuracy on the test nodes or all nodes?

From your Table 3. the 64-layer GCN, the test accuracy is already 28.7. However, from your Figure 1, the accuracy is far from 20. The 16-layer and 8-layer GCN also have the same problem. Can you tell me why this happens. Do I missing something?

It would be better if you can share the GCN code for plotting your Figure 1

Thank you very much

[chennnM]

The data split in Figure 1 is different from that in table 3. In Figure 1, we randomly select about 100 nodes for training and plot the figure on other nodes. The code will be released soon.