Mastering the Game of Gobang from Scratch

Objective

In this work, we investigate the game of Gobang using multiple algorithms, including Mini-Max Tree Search with Alpha-Beta Pruning, Monte Carlo Tree Search, Genetic Algorithm, Deep Q-Network, and AlphaZero.

Accomplishments

• Implemented Minimax Algorithm with Alpha-beta pruning
• Implemented Genetic Algorithm
• Implemented progressive MCTS
• Trained and Implemented DQN-based agent
• Trained and Implemented AlphaZero
• Compared the Algorithms across Different Board Size and Training Iterations

Results

We found that the minimax algorithm with alpha-beta pruning and Monte Carlo tree search worked well on smaller board sizes, while AlphaGoZero outperformed other algorithms after 3,000 epochs of training.