AlphaGo Knock-off

This is a knock-off of the AlphaGo algorithm, which is a deep learning algorithm that can play the game of Go. The algorithm is described in the paper Mastering the Game of Go with Deep Neural Networks and Tree Search by Silver et al. The algorithm is implemented in Python using the Keras deep learning library.

Authors

• Seran Shanmugathas
• Ferdinand T. Eide

Requirements

• Python 3.5
• Keras 2.0.6
• Tensorflow 1.1.0
• Numpy 1.12.1

Usage

Training

To train the model, run the following command:

python3 train.py

The training data is stored in the data directory. The model is saved in the model directory.

Playing

To play against the model, run the following command:

python3 main.py

The model is loaded from the model directory.

Folder structure

alphaGo-knock-off/
├── game/
│ ├── __init__.py
│ ├── board.py
│ ├── moves.py
│ └── win_conditions.py
├── neural_network/
│ ├── __init__.py
│ ├── anet.py
│ └── training.py
├── mcts/
│ ├── __init__.py
│ ├── node.py
│ └── search.py
├── reinforcement_learning/
│ ├── __init__.py
│ ├── actor.py
│ ├── critic.py
│ ├── environment.py
│ └── training.py
├── topp/
│ ├── __init__.py
│ ├── competition.py
│ ├── evaluation.py
│ └── ranking.py
├── main.py
└── requirements.txt