/cfr-and-her-variant

Realization of counterfactual regret minimization basic method and related variants.

Primary LanguagePython

cfr-and-her-variant

CFR(Counterfactual Regret Minimization) algorithm is a important method for solving imperfect information game, and proved to be workable relative to DRL. This code repo about realization of CFR basic method and related variants.

already done

  1. regret matching

pancake

  • CFR-raw(basic cfr algorithm)
  • Monte Carlo CFR
  • CFR+
  • Linear CFR
  • Deep CFR

Actually, I may code other imperfect information methods such as NFSP、RPG... sometime

References

[1] Tammelin O. Solving large imperfect information games using CFR+[J]. arXiv preprint arXiv:1407.5042, 2014.

[2] Brown N, Sandholm T. Safe and nested subgame solving for imperfect-information games[C]//Advances in neural information processing systems. 2017: 689-699.

[3] Brown N, Lerer A, Gross S, et al. Deep Counterfactual Regret Minimization[C]//International Conference on Machine Learning. 2019: 793-802.

[4] Brown N, Sandholm T. Superhuman AI for multiplayer poker[J]. Science, 2019: eaay2400.

[5] Burch N, Johanson M, Bowling M. Solving imperfect information games using decomposition[C]//Twenty-Eighth AAAI Conference on Artificial Intelligence. 2014.

[6] Moravčík M, Schmid M, Burch N, et al. Deepstack: Expert-level artificial intelligence in heads-up no-limit poker[J]. Science, 2017, 356(6337): 508-513.