Onitama, the game

Final project - IASD - Monte Carlo Tree Search

We implement the game Onitama in Python and test different MCTS algorithms to play the game.

The script onitama.py contains the definition of the game with its rules and evolution. The class Board represents a game, and the class Move represents a move.

The script play_functions.py contains the different algorithms that play the game. The idea is that a player has to be general enough to play any other game. The standard arguments of a player are a transposition table (\modules\transposition_table.py:T_Table) that stores move statistics and other game constants passed from the game class, and a board representing the state of the game. Players must return a move given a transposition table, a board, and any other needed parameter.

Use GUI to play the game

We made a simple implementation of the game using Pygame. If you want to play against one of the algorithms, launch the script GUI_oni.py. The parameters you can pass to this script are Run python GUI_oni.py --help for all the details.

  --enemy    Select the enemy AI to play against from the options available (use --help to see them)
  --n        Budget for the AI
  --play_as  red or blue (red plays first)

Compare the different algorithms

The script modules\tournament.py is designed to run a tournament between the bots. To run a tournament, you must create a JSON file containing the details for the tournament including its name, the bots included and their paramteres. See the bot_fights folder for an example. The results are saved in the data folder by default, but you can pass the desired path as argument. Run python tournament.py --help for all the details.

Create your own player

We tried to create general classes so that it is easy to implement your own AI. See the play_functions script to see how we developped the players. The idea is to create a function that receives at least two inputs: a transposition table T_Table used to store move statistics and other variables from the game, and a board which contains all the game logic and state. You can also recieve other arguments if needed. This function must return a Move object representing the move to play next. Add the bot to the bot dictionary in play_functions for it to be available for the GUI.