Skip to content
/ snake-ai Public
forked from linyiLYi/snake-ai

An AI agent that beats the classic game "Snake".

License

Notifications You must be signed in to change notification settings

OFA97/snake-ai

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

29 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SnakeAI

简体中文 | English

This project contains the program scripts for the classic game "Snake" and an artificial intelligence agent that can play the game automatically. The intelligent agent is trained using deep reinforcement learning and includes two versions: an agent based on a Multi-Layer Perceptron (MLP) and an agent based on a Convolution Neural Network (CNN), with the latter having a higher average game score.

File Structure

├───main
│   ├───logs
│   ├───trained_models_cnn
│   ├───trained_models_mlp
│   └───scripts
├───utils
│   └───scripts

The main code folder for the project is main/. It contains logs/, which includes terminal text and data curves of the training process (viewable using Tensorboard); trained_models_cnn/ and trained_models_mlp/ respectively contain the model weight files for the convolutional network and perceptron models at different stages, which can be used for running tests in test_cnn.py and test_mlp.py to observe the actual game performance of the two intelligent agents at different training stages.

The other folder utils/ includes two utility scripts. check_gpu_status/ is used to check if the GPU can be called by PyTorch; compress_code.py can remove all indentation and line breaks from the code, turning it into a tightly arranged single line of text for easier communication with GPT-4 when asking for code suggestions (GPT-4's understanding of code is far superior to humans and doesn't require indentation, line breaks, etc.).

Running Guide

This project is based on the Python programming language and mainly uses external code libraries such as PygameOpenAI GymStable-Baselines3. The Python version used for running the program is 3.8.16. It is recommended to use Anaconda to configure the Python environment. The following setup process has been tested on the Windows 11 system. The following commands are for the console/terminal (Console/Terminal/Shell).

Environment Configuration

# Create a conda environment named SnakeAI with Python version 3.8.16
conda create -n SnakeAI python=3.8.16
conda activate SnakeAI

# [Optional] To use GPU for training, manually install the full version of PyTorch
conda install pytorch=2.0.0 torchvision pytorch-cuda=11.8 -c pytorch -c nvidia

# [Optional] Run the script to test if PyTorch can successfully call the GPU
python .\utils\check_gpu_status.py

# Install external code libraries
pip install -r requirements.txt

Running Tests

The main/ folder of the project contains the program scripts for the classic game "Snake", based on the Pygame code library. You can directly run the following command to play the game:

cd [parent folder of the project]/snake-ai/main
python .\snake_game.py

After completing the environment configuration, you can run test_cnn.py or test_mlp.py in the main/ folder to test and observe the actual performance of the two intelligent agents at different training stages.

cd [parent folder of the project]/snake-ai/main
python test_cnn.py
python test_mlp.py

Model weight files are stored in the main/trained_models_cnn/ and main/trained_models_mlp/ folders. Both test scripts call the trained models by default. If you want to observe the AI performance at different training stages, you can modify the MODEL_PATH variable in the test scripts to point to the file path of other models.

Training Models

If you need to retrain the models, you can run train_cnn.py or train_mlp.py in the main/ folder.

cd [parent folder of the project]/snake-ai/main
python train_cnn.py
python train_mlp.py

Viewing Curves

The project includes Tensorboard curve graphs of the training process. You can use Tensorboard to view detailed data. It is recommended to use the integrated Tensorboard plugin in VSCode for direct viewing, or you can use the traditional method:

cd [parent folder of the project]/snake-ai/main
tensorboard --logdir=logs/

Open the default Tensorboard service address http://localhost:6006/ in your browser to view the interactive curve graphs of the training process.

Acknowledgements

The external code libraries used in this project include PygameOpenAI GymStable-Baselines3. Thanks all the software developers for their selfless dedication to the open-source community!

The convolutional neural network used in this project is from the Nature paper:

[1] Human-level control through deep reinforcement learning

About

An AI agent that beats the classic game "Snake".

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%