Question about unsupervised_Cora_Citeseer

Shen-Lab / GraphCL

[NeurIPS 2020] "Graph Contrastive Learning with Augmentations" by Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, Yang Shen

MIT License

559 stars 103 forks source link

Question about unsupervised_Cora_Citeseer #49

Open Zhen-Tan-dmml opened 2 years ago

Zhen-Tan-dmml commented 2 years ago

Hi @yyou1996, thanks for your great work. Just have a question about the GraphCL for node classification in unsupervised_Cora_Citeseer. The default batch_size is set to 1. How do you generate negative pairs?

AlexWanghaoming commented 2 years ago

Hi @yyou1996, it seems that batch sizes make no sense in this task. Could please provide a version considering batch size. Thanks.

yyou1996 commented 2 years ago

Thanks for the question.

I think for cora/citeseer the common practice is full-batch training, that though bs=1, it is actually equal to the node number. Whilst, I need some confirms here. @yongduosui