search

zhumeiqiBUPT / AM-GCN

AM-GCN: Adaptive Multi-channel Graph Convolutional Networks
MIT License
219 stars 46 forks source link

Code issues #10

Closed xnuohz closed 3 years ago

xnuohz commented 3 years ago

Hi, I am a little confused about this line, could you explain it? And it seems that this line doesn't influence the result if I remove it. Thanks:)

zhumeiqiBUPT commented 3 years ago

Hello, these two lines are used for building symmetric adjacency matrices, which are the same with that in pygcn. Since our input sadj are already symmetric adjacency matrices, it doesn't influence the result.

xnuohz commented 3 years ago

Thanks for your reply. KNN graph should be symmetric too, so it also doesn't influence the result if this line is removed. But accuracy is 72.0 instead of 73.3 when I run python main.py -d citeseer -l 20, do I misunderstand something?

zhumeiqiBUPT commented 3 years ago

Please note that edge files in folder "/data/dataname/knn/" for KNN graph just write each edge for one time, which means the input graph is not symmetric before processing. The edge files in folder "/data/dataname/XX.edge" write each edge for two times, so the input graph is symmetric. Anyway, you can always keep these two lines just in case.