MCTS - tic-tac-toe

tic-tac-toe Monte Carlo Tree Search

Instsallation

How to install package depenencies.

Prerequisties

• Python >= 3.7

Install Python packages

pip install -r requirements.txt

Available initial states

• empty game
• initial state (--ini-game assignment):

Usage

• --n-rollouts is number of rollouts per Monte Carlo Tree Search interation (aka computational budget)
• --ini-game empty or assignment
• --verbose optional flag to increase number of logs

python  main.py --n-rollouts N_ROLLOUTS --ini-game {empty,assignment} [--verbose]

For example: bash

python main.py --n-rollouts 10 --ini-game assignment

Possible output

x plays with MCTS
Initial game:
_ o x
_ x _
o _ _
Player 'x' moved to (1,2)
_ o x
_ x x
o _ _
Player 'o' moved to (2,1)
_ o x
_ x x
o o _
Player 'x' moved to (1,0)
_ o x
x x x
o o _
Winner: x

Bulk testing

To run the script N times you can use the test.py and plot.py scripts.

• test.py runs main.py with dirrerent parameters and pickles results
• plot.py loads test results and create plots

Usage

python test.py
python plot.py

Convergence

Probabilities and Q-values in initial states

Empty game

Assignemnt game