AWAC: cannot reproduce result in the paper

rail-berkeley / rlkit

Collection of reinforcement learning algorithms

MIT License

2.43k stars 547 forks source link

AWAC: cannot reproduce result in the paper #136

Open bmbyzy opened 3 years ago

bmbyzy commented 3 years ago

I ran the AWAC code and plotted the output 'progress.csv' file for HalfCheetah as below

The return only matches the online training part shown in the paper, but fails to reproduce the return obtained from offline training. May I have your suggestions on how to reproduce the same plot in the paper? Thanks.