search

thu-ml / tianshou

An elegant PyTorch deep reinforcement learning library.
https://tianshou.org
MIT License
7.8k stars 1.12k forks source link

Does the current PPO support discrete actions? #101

Closed minruixu closed 4 years ago

minruixu commented 4 years ago

Hi, how an exciting work you have done in tianshou! But there still some doubts while I use the code in my experiments. I have found that there are no different in the continues version PPO and discrete version PPO. However, your experimental results show that the discrete version PPO work well in CartPole-v0. But when I change the env to MountainCar and Acrobot it does not work at all.

Trinkle23897 commented 4 years ago

Yes but both MountainCar and Acrobot need another set of hyperparameters. We are working on it. We will release something like stable-baselines-zoo: https://github.com/araffin/rl-baselines-zoo, it's on our roadmap.

Trinkle23897 commented 4 years ago

Hi, please use the newest version to run it. I think it works well.

155