why the cl_loss can be a negative number on the tranferlearning dataset chem?

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

548 stars 103 forks source link

why the cl_loss can be a negative number on the tranferlearning dataset chem? #23

Closed check-777 closed 3 years ago

check-777 commented 3 years ago

ha-lins commented 3 years ago

Thanks for your help. I wonder if this problem is fixed and reproduce the pretraining performance successfully. Are the downstream performance close to the paper results for the latest version of pyG? Hope to chat more about this via WeChat (ls825521138).

peizhenbai commented 2 years ago

Hi @ha-lins @check-777,

Hope you are doing well. I also meet the same issue while reproducing transfer learning result. May I ask what the reason is and whether it is fixed in your implementation? Thank you very much.

yyou1996 commented 2 years ago

Hi @pz-white,

Please refer to another issue per my answer https://github.com/Shen-Lab/GraphCL/issues/50.

peizhenbai commented 2 years ago

Hi @pz-white,

Please refer to another issue per my answer #50.

@yyou1996 Thank you so much, Yuning! That's very helpful to me :-)

Hope to have a nice day!