1jsingh / alphazero_tictactoe

Implementation of alphazero learning algorithm for 6-6-4 Tictactoe

Alphazero for 6-6-4 Tictactoe

Implementation of alphazero learning algorithm for 6-6-4 Tictactoe

Trained Alphazero agent vs human!

The trained alphazero agent easily defeated me!

• Player1 (Red) : Alphazero
• Player2 (Blue) : Me

Project Structure

• alphazero_tictactoe.ipynb: Jupyter notebook for training alphazero for normal 3x3x3 tictactoe
• alphazero_tictactoe_advanced.ipynb: Jupyter notebook for training alphazero for complex 6x6x4 tictactoe
• MCTS.py: Helper file for Monte Carlo Tree Search
• Play.py: File for helping with interactive game play
• ConnectN.py: File defining the game structure
• 6-6-4-pie.policy: Trained policy weights for 6x6x4 tictactoe

Instructions

• Follow the step by step instructions in alphazero_tictactoe_advanced.ipynb to train your own alphazero agent!