A question: LSTM + PPO - Githubissues

thu-ml / tianshou

An elegant PyTorch deep reinforcement learning library.

https://tianshou.org

MIT License

7.75k stars 1.12k forks source link

A question: LSTM + PPO #498

Closed tesla-cat closed 2 weeks ago

tesla-cat commented 2 years ago

A question

I want to build a small scale version of Open AI Five
And I learnt that it uses LSTM + PPO
suppose I build a network model using LSTM, then should I use this network for both Actor and Critic?
if not, what should I use?
I have tried to find online tutorials about using RNN as model for RL, but didn't find useful ones
can you recommend some hands-on tutorials on this?
it would be even better if you can create a tutorial on this using tianshou, thank you!

Trinkle23897 commented 2 years ago

https://tianshou.readthedocs.io/en/master/tutorials/cheatsheet.html#rnn-style-training https://github.com/thu-ml/tianshou/issues/486#issuecomment-1002665193

vwxyzjn commented 2 years ago

Hey @tesla-cat, coincidentally we have been building OpenAI Five style bots in gym-microrts, which looks like

We have successfully prototyped PPO + LSTM in the Atari games, and we hope to prototype PPO + LSTM in gym-microrts as well. If you are interested in helping with this prototype, please let me know :)

thu-ml / tianshou

A question: LSTM + PPO #498

A question

if not, what should I use?