Hyperparameters used for training

[grach0v]

Hello,
I tried to replicate the results in the paper using default parameters set in the repository and I didn't get any meaningful training.
So I assume some other hyperparameters were used for training.
Could you please share them, so we can reproduce the results.
Thanks.

[xukechun]

Hi, so sorry for the late reply! Maybe you can try our conda environment. Based on the previous discussion, it appears that some version of packages will be relevant during both training and testing. Or we can publish a more stable version for RL training later.