Need help with snake-ga?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

370 Stars 791 Forks 30 Commits 9 Opened issues


AI Agent that learns how to play Snake with Deep Q-Learning

Services available


Need anything else?

Contributors list

# 94,642
26 commits
# 324,791
1 commit

Deep Reinforcement Learning

Project: Train AI to play Snake


This project has been recently updated and improved: - It is now possible to optimize the Deep RL approach using Bayesian Optimization. - The code of Deep Reinforcement Learning was ported from Keras/TF to Pytorch. To see the old version of the code in Keras/TF, please refer to this repository: snake-ga-tf.


The goal of this project is to develop an AI Bot able to learn how to play the popular game Snake from scratch. In order to do it, I implemented a Deep Reinforcement Learning algorithm. This approach consists in giving the system parameters related to its state, and a positive or negative reward based on its actions. No rules about the game are given, and initially the Bot has no information on what it needs to do. The goal for the system is to figure it out and elaborate a strategy to maximize the score - or the reward. \ We are going to see how a Deep Q-Learning algorithm learns how to play Snake, scoring up to 50 points and showing a solid strategy after only 5 minutes of training. \ Additionally, it is possible to run the Bayesian Optimization method to find the optimal parameters of the Deep neural network, as well as some parameters of the Deep RL approach.


This project requires Python 3.6 with the pygame library installed, as well as Pytorch. If you encounter any error with

, you might need to install Visual C++ 2015-2019 (or simply downgrade your pytorch version, it should be fine). \ The full list of requirements is in
git clone [email protected]:maurock/snake-ga.git


To run and show the game, executes in the snake-ga folder:


Arguments description:

  • --display - Type bool, default True, display or not game view
  • --speed - Type integer, default 50, game speed

The default configuration loads the file weights/weights.h5 and runs a test.

To train the agent, set in the file -

params['train'] = True
The parameters of the Deep neural network can be changed in by modifying the dictionary
in the function

If you run from the command line, you can set the arguments

. This way, the game display is not shown and the training phase is faster.

Optimize Deep RL with Bayesian Optimization

To optimize the Deep neural network and additional parameters, run:

python --bayesianopt=True

This method uses Bayesian optimization to optimize some parameters of Deep RL. The parameters and the features' search space can be modified in by editing the

dictionary in

For Mac users

It seems there is a OSX specific problem, since many users cannot see the game running. To fix this problem, in update_screen(), add this line.

def update_screen():
pygame.event.get() #

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.