Tried new approach for judge_chow as suggested in https://github.com/datamllab/rlcard/issues/211 . The performance seemed to increate just a little but not significant enough to solve our issues. Our assumption is that the function judge_chow() in the judger class isn't the main problem. A possible bottleneck for mahjong is judge_hu in judger class. We'd appreciate any suggestions on how to simplify that part.
Tried new approach for judge_chow as suggested in https://github.com/datamllab/rlcard/issues/211 . The performance seemed to increate just a little but not significant enough to solve our issues. Our assumption is that the function judge_chow() in the judger class isn't the main problem. A possible bottleneck for mahjong is judge_hu in judger class. We'd appreciate any suggestions on how to simplify that part.