This project contains best effort open source implementation of a poker AI using several methods. The AI models are built by simple rule-based method + simulations, Monte-Carlo tree search and opponent modeling. The game is built mainly on terminal, but we also have a pygame expansion for playing against bots.

This repository assumes Python 3.11 or newer is used. Used package:

• bext
• colorama
• pygame

Clone the repo to your own machine:

git clone https://github.com/Sylviss/Poker_AI_Project

Firstly, you should change some constant in poker_ai/constant.py. These are the default settings, which should be enough for you to test with the AI:

PLAYER = 2
# The number of players
PREFLOP_BIG_BLIND = 10
# Value of the big blind pre-bet.
INIT_MONEY = 50 * PREFLOP_BIG_BLIND
# The starting money of all players
STOP = 0
# 0 for stop after every game, 1 to skip stop
INDICATOR = 1
# 0 is for testing against all human-controlled
# 1 is for human vs bots: Player 1 will be human, all others will be bot
# 2 is all bots for testing purpose
MULTIPROCESS = 1
# 0 is for single-processing, slower. You should use this if you encounter some problems with multiprocessing.
# 1 is for multi-processing, faster and recommended.
MODEL=7
# Choose AI module for all bot in a normal game for main.py:
# -1: enumeration ai agent
# 0: first approach mcs ai agent
# 1: second approach mcs ai agent
# 2: mcts ai agent
# 3: all in ai agent
# 4: super random ai agent
# 5: first approach mcs ai agent with opponent modeling
# 6: second approach mcs ai agent with opponent modeling
# 7: mcts ai agent with opponent modeling
# Note that opponent modeling is implemented with multiprocess, so don't use opponent modeling if you encounter some problems with multiprocessing.

Running main.py will run the game in terminal with all of the above settings:

python main.py

If you want to play the game in a fancy pygame implement, run fancy_table.py:

python fancy_table.py

Note that:

• The pygame implement only support games of player vs AIs. It currently do not support player vs players and AI vs AIs, since it's impossible for players to play in 1 machine and AI games are faster on terminal.
• The game logs and many useful informations are printed out in the terminal.

Firstly, you should make a backup file of poker_ai/ai/ml/default_data.json, as it is our model's default data. To start off, delete/rename the default_data.json. Run training.py:

python training.py #mode

All of the modes for traning:

• 1: Best suited for model againts 2 players.
• 2: Best suited for model againts 3-4 players.
• 3: Best suited for model againts 5-6 players.
• 4: Best suited for model againts 2-6 players. Note that the opponent modeling is only useful with enough training time, with the minimum should be 2-3 hours. If you don't have much time, please use our pre-trained model.

Edit tester.py:

if __name__=="__main__":
    game_loop_model_test(num_players,init_money,games,models_list)

Changes the parameters in game_loop_model_test:

• num_players: number of player in testing
• init_money: the amount of starting money
• games: the number of games to test
• models_list: the models that you want to use in the test. The first element will be Player 1's model, second will be Player 2's model and so on. Afer that, run tester.py:

python tester.py

After testing, the file will print out the number of games win by every models.

• The multiprocessing has a strange interaction with KeyboardInterrupt. As such, if not necessary, don't try to cancel any running programs with multiprocessing running. The best bet to cancel any programs is when it's your turn to act in a player vs AI mode.
• All of the materials and references we used in this project is inside books & materials folders.
• If there's any major problems with the project, please contact with me through Teams. I will fix the problems as soon as possible.