reproduction of kseg

[DopamineLcy]

Thank you for your impressive work!
I tried reproducing your results of kseg strictly according to your README without changing any config, however, I surprisingly found the results of kseg notably lower than the reported ones.

Is there any potential mistake that leads to the inconsistency?
Thank you again and looking forward to your reply.

[DopamineLcy]

Would you mind sharing the data splits of keg, I'm not sure whether the stochasticity of glob leads to different data plits.

[lin-tianyu]

Hi, thanks for your interest. Unfortunately, I don't have a data splits of kseg, since the data split I use is in the dataloader file kseg.py with the _parse_data_list method. I am not sure whether this is a problem from glob because my previous experiment results seems consistent.

Have you checked the tensorboard of the training log? What does it look like? Maybe you can get more insight from the tensorboard.

If you still cannot fix the problem, I will try reomplement it when I'm free. Thanks!