search

microsoft / med-deadend

Code for the Medical Deadend Paper at NeurIPS 2021
MIT License
47 stars 18 forks source link

Would you please explain some about learning rate #5

Closed yss858 closed 2 years ago

yss858 commented 2 years ago

Hello I see in your paper that the learning rate of the best SC-network is 5e-4 and epochs are 600. Then I see that the learning rate of the config.yaml file is 1e-4 and epochs are 400. Would you please let me know whether you use the former or the latter in the experiment? Thank you！

twkillian commented 2 years ago

So, after conferring with @fatemi, we've realized that our paper has this slight error. All experiments in the paper were produced by the code in this repository. So to accurately replicate what we've done, use the configuration file as currently provided.