search

QueuQ / CGLB

Other
49 stars 14 forks source link

[Reproduce] Hyperparameters #2

Closed doviettung96 closed 1 year ago

doviettung96 commented 1 year ago

Dear @QueuQ , I find this work interesting and helpful for continual learning with graphs. According to the paper, the results from ERGNN is quite dominant. However, I can't reproduce your results with the default args. Thus, I would like to know the set of parameters that you used. Thanks.

doviettung96 commented 1 year ago

Ops. I found the hyperparameters in the Appendix. Even though it lacks the number of training epochs for NGCL, I tried with 100 epochs and it seems to work. Thanks.

QueuQ commented 1 year ago

Thanks for mentioning this issue, the default number of epochs is 200, I should have set it as 200 in the code, I have just updated it