datamllab / rlcard

Reinforcement Learning / AI Bots in Card (Poker) Games - Blackjack, Leduc, Texas, DouDizhu, Mahjong, UNO.
http://www.rlcard.org
MIT License
2.87k stars 619 forks source link

how to train mahjong game? #298

Closed jacy closed 5 months ago

daochenzha commented 1 year ago

@jacy Just need to change the env to mahjong. Note that the state and action features of Mahjong are not well-engineered. So to achieve a better performance, it is recommended to spend more effort on the feature design.

jacy commented 1 year ago

hi @daochenzha, is it just to use the random model to train?

daochenzha commented 1 year ago

@jacy Yes, you should be able to see reward improvement with a random model. But it could not be satisfactory because the features are not well-engineered.

jacy commented 1 year ago

ahh ok, seems it might still need to take lots of effort to train it reach to a normal human level

jacy commented 8 months ago

have been training mahjong game using DMC, seems the rewards didn't improve even train for several days, anyone can shed some light?