Alpha Gomoku Group Project

Written and maintained by Matthew Boakes, Harry Clarke, Matthew Clayton, Max Harris, and Jamie Pont.

How to install (on a Linux machine connected to the Kent vpn)

• clone the directory from git
• install python3
• pip install -I tensorflow==0.12.1
• run sudo sh ./setup.sh

How to install (Windows, Mac, Or a Linux machine not connected to the Kent vpn)

• Same as Linux (on the Kent vpn) but instead of running setup.sh command, create a folder in the root directory (the one with src in) and create a the folder resources
• Download training.zip from the following link:
• https://drive.google.com/file/d/0B86_99L1GbtLU3BybVB6UEZfakE/view?usp=sharing
• Unzip the training file from training.zip into the resources folder

How to run

• To train the neural network:
• python3 neural_network_main.py
• To run it with previously trained weights pass in the following parameter to the above command: true
• To play gomoku against the trained neural network and tree search:
• python3 gomoku_main.py

Introduction

The AlphaGomoku project aims to bring the concept behind Deep Mind's AlphaGo ("Mastering the game of Go with deep neural networks and tree search", David Silver et al., 28/01/16) to a game called "Gomoku".

Project Goals

• September
• Finish learning from the book
• October
• Have a basic neural netowkr put together for our understanding
• Finish Gomoku App for playing the game
• End of Christmas
• Apply knowledge gained from basic NN project and develop a NN around gomoku
• After Christmas
• Fine tune/finish project