/Five-Card-Stud

Primary LanguagePython

Five Card Stud

Online Demo (currently only support two players, one human player and one AI player)

Try it out: Five Card Stud Online Demo

Test the game locally

Make sure you are in the master branch

Run the game simulation locally

  • Clone the repository

    git clone https://github.com/Leheng620/Five-Card-Stud.git

  • Install dependencies:

    pip install -r requirements.txt

  • At the root directory, run

    python five_card_stud/game.py <option> <num_of_game_play>

    Replace <option> with the following:

    • To run simulation between Random agent and Uniform agent:

      <option>: random_vs_uniform

    • To run simulation between MCTS agent and Uniform agent:

      <option>: mcts_vs_uniform <iteration>, <iteration> is the number of iterations the MCTS agents will run. Minimum 100.

    • To run simulation between MCTS agent and Random agent:

      <option>: mcts_vs_random <iteration>

    • To run simulation between MCTS agent and MCTS agent:

      <option>: mcts_vs_mcts <iteration> <iteration> The first <iteration> is the number of iterations the first MCTS agent will run. The second <iteration> is the number of iterations the second MCTS agent will run.

    <num_of_game_play> is the number of games to simulate. A game ends when an agent runs out of balance.

    After the simulation ends, a counter object will be printed out. The player 0 is the player before "vs" in <option>, player 1 is the player after "vs" in <option>

Play games with a MCTS agent (Our best agent so far) in command line. You can also try the online demo

python five_card_stud/test_human.py

Run tests

test_efficiency.py contains the script to test algorithm running time, the argument is the same as game.py except that we fix the number of game plays to be 100.

test_mcts_iter.py tests the MCTS agent with different iteration numbers.

test_mcts_ucb.py tries different values of the constant c in the UCB score in MCTS algorithm.

Concepts

Agent

Uniform agent

Takes available actions randomly with no bias.

Random agent

Takes available actions randomly based on weighted probability. The weight is the heuristics.

MCTS agent

An agent that is equipped with Monte Carlo Tree Search algorithm to take the best actions given the current state of the game.

Environment

Defined in gameState.py. Currently there can be two players only, and the card deck only contains 26 cards from 8 to A.

Simulation Result

  • Random Agent vs Uniform Agent

    • Run with command python five_card_stud/game.py random_vs_uniform 1000
    • Result: Counter({0: 896, 1: 104})

  • MCTS Agent with iteration == 100 vs Uniform Agent

    • Run with command python five_card_stud/game.py mcts_vs_uniform 100 1000
    • Result: Counter({0: 913, 1: 87})

  • MCTS Agent with iteration == 100 vs Random Agent

    • Run with command python five_card_stud/game.py mcts_vs_random 100 1000
    • Result: Counter({0: 601, 1: 399})

  • MCTS Agent with iteration == 100 vs MCTS Agent with iteration == 1000

    • Run with command python five_card_stud/game.py mcts_vs_mcts 100 1000 1000
    • Result: Counter({0: 505, 1: 495})