Snake-AI

Train, manage, and control your own AI snakes playing an incredibly classic game! All components are made completely from scratch including the game logic, AI algorithms, and genetic training. The training employs multiprocessing for speed. I designed the GUI and rendered it with Pygame (pygame.org). Read below to learn about the different types of AI.

Get started:

• Download Python 3.8+ https://www.python.org/downloads/
• Clone repository git clone https://github.com/gholmes829/Snake-AI.git
• Install dependencies python3 -m pip install -r requirements.txt
• Run with python3 __main__.py

Note: Windows users may need to run commands with python instead of python3

Command Line Arguments:

• Classic player control python3 __main__.py -player
• Select an AI to play python3 __main__.py -ai
• Watch a saved game python3 __main__.py -saved
• Train AI snakes python3 __main__.py -train

Types of AI:

• Neural Network -- uses a deep neural network to output left, straight, or right
• Pathfinding -- uses an A* algorithm with backup floodfill to find direct paths to food
• Floodfill -- moves towards food but always tries to move in direction resulting in most open spaces
• Cycle -- calculates hamiltonian cycle approximation for path
• Multi -- specialized neural network that chooses between pathfinding, cycle, and neural network as game progresses
• Hierarchical -- specialized neural network that chooses between three different neural networks as game progresses

Notes:

• Change settings in core\settings.py including map size, target frame rate, default architectures and more
• Cycle AI can consistently win game if map size is square, even, and snake spawns in top row horizontally
• Hierarchical is the best overall AI in terms of both performance and consistency
• My personal best score is like 25 or something which is why I needed to make the AI