Play Snake with Deep Q-Learning

by Jae-Hyeong, Sim(@Kuax-Meat)

If you have any question, suggest an issue or send an email to me. mirelurk at kuax.org

1. Introduction

This repo is an agent that plays snake game with Google DeepMind's DQN(NIPS 2013).

2. Dependencies

Fully compatible with

• Windows 10 Professional 64-bit
• Anaconda 4.2.0 64-bit (Python 3.5.x)
• Tensorflow RC 1.0
• Pygame
• OpenCV

3. Convolutional Neural Network

Uses 3 hidden Convolutional NN layer(same as DQN Nature 2015).

4. How to run

simply type this command.

> python play.py

and if you want to train,

> python train.py

If you want to fresh start, comment this line(insert # on front of the line) and type

> python train.py

5. Result

Details of Snake for 8 hours(1,383,950 Frames and 69,989 Episodes) with Nvidia Geforce GTX 1070

6. Disclaimer

This repo is highly based on

• https://github.com/devsisters/DQN-tensorflow
• https://github.com/asrivat1/DeepLearningVideoGames

and yes, Google DeepMind's DQN. https://deepmind.com/research/dqn/

Game Snaky Raw Code from https://inventwithpython.com/pygame/chapter6.html

Thanks to Sung Kim(@hunkim)

http://hunkim.github.io/ml/