- Too conservative
- Can't proactively beat benchmark
- Others use neutral network/ contraction
- Take-away1: actively test against benchmark
- Take-away2: actively test against other player
- Tons of fun though we fail
I have read through at least 5 github repo, and some paper related to Texas Holdem. No limit Texas Holdem, in the terms of reinforcement learning, is multi-agent imperfect information game with 10^161 state space and 10^4 action space.
AI repo
Must-check Awesome-Game-AI -- a gathering of all poker AI resourecs
Rlcard -- it's an environment, don't see anything of pariticular interest
DeepHoldem -- take 12s in flop and 7s in turn round, also it's written in Lua
OpenAI Poker gym -- it's a tournament environment
Deep mind pokerbot -- This one looks promising, espeically the MonteCarlo_numpy.py (calculate winning probability) and DecisionMaker.py (decide the action) under Poker/poker/decisionmaker may be quite helpful.
Links list below
Awesome-Game-AI
Rlcard
Rlcard-paper
Open AI gym
DeepHoldem
Deep mind pokerbot
九坤提供的代码模型，我们的代码改动需要在写在UbiQuant文件夹。同时UbiQuant里面有自己的readme
source root of UbiQuant:
UbiQuant
UbiQuant/modules/texaspoker
(两个都要设为source root)
利用模特卡罗模拟计算胜率的代码，也有自己的readme
source root of WinRatioCalculator:
WinRatioCalculator
clubs (c)