rlcard
rlcard copied to clipboard
datamllab
Reinforcement Learning / AI Bots in Card (Poker) Games - Blackjack, Leduc, Texas, DouDizhu, Mahjong, UNO.
Here are some games that are similar to Dou Dizhu and Uno that may be interesting - Zheng Shangyou https://en.wikipedia.org/wiki/Winner_(card_game) - Japanese Daifugō https://en.wikipedia.org/wiki/Daifug%C5%8D - Russian Durak https://en.wikipedia.org/wiki/Durak - Baltic...
Rummy is still very poplar in Europe and America, it might be a good idea to include one of these variants into your system. A side note: Rummy is "the...
When I set self.init_chips=2000 (1000bbs) in game.py, the code runs extremly slow. I found out it's because of the action space setting. The num of leagl actions is too large(from...
how to make a baccarat game with the rlcard framework?
Im part of the dev team on PettingZoo. We were through our CI trying to reduce the number of warnings and spotted this warning ``` test/pytest_runner_test.py: 99 warnings test/unwrapped_test.py: 6...
For below cards, judge_hu return false, but it should be true: Card('dots', '4'), Card('dots', '4'), Card('dots', '4'), Card('characters', '2'), Card('characters', '2'), Card('characters', '2'), Card('characters', '2'), Card('characters', '3'), Card('characters', '4'), Card('bamboo',...
在 evn 中 run 函数内部，step 所执行的是单牌桌的单步操作，跟牌时的可操作数量往往很少（单牌只能根单牌和炸弹），所以频繁的调用 AI 计算次数的同时，计算量并不高。整体来说 GPU 的使用率会很高，多进程（CPU 核数多）时 GPU 可能会打满。 而使用多牌桌模型，将多个牌桌的 setp 所需数据，组合一次性送入 AI 模型计算，然后拆分到每桌进行逻辑处理。 计算量可能不会减少，但是送入的次数会大大降低。可以多尝试一些牌桌数量，这里测试的结果是，相比单桌 GPU 运算慢载（CPU 核数多，所以 GPU 成了瓶颈），修改成 32 桌同时进行，GPU 占用率会下降到百分之十几，帧率可以提高数倍（CPU 会成为瓶颈）。更多的牌桌数，会导致 GPU 的运算量下降到个位数，显存有一定提升，但不明显。更多的牌桌数不意味着更高的帧率，有个平衡点，这里测试的是...
