search

openai / baselines

OpenAI Baselines: high-quality implementations of reinforcement learning algorithms
MIT License
15.69k stars 4.86k forks source link

can i apply any baseline algo to game like chess? #1085

Open Unimax opened 4 years ago

Unimax commented 4 years ago

hi, i was wondering if i can apply the baselines algo on the game chess either https://github.com/genyrosk/gym-chess or a custom chess env. based on python-chess.

i was thinking 18x8x8 input and 64*64+permotion as action space.

let me know if anyone can help me do this. Note: goal is not to make the best chess ai in the world :)

Hadjubuntu commented 4 years ago

I guess you can try DQN algorithm but traditionnal AI/RL algorithms for chess are based or inspired by AlphaZero: By combining MCTS and Q-Value estimator, you may find the results from the MuZero paper: https://arxiv.org/abs/1911.08265

Note that Ray framework have released an "AlphaZero" like algorithm: https://github.com/ray-project/ray/tree/master/rllib/contrib/alpha_zero

Best regards, Adrien

Unimax commented 4 years ago

thanks, @Hadjubuntu. I knew about AlphaZero and MuZero but i wanted to compare the strength of algos like PPO,SAC and DQN on the game like chess (basically something more useful/interesting than standard gym envs). Also, thanks for the Ray framework link didn't knew that project.