https://github.com/nikorasu/snakeai

WIP project to use machine learning to play Snake!
https://github.com/nikorasu/snakeai

ai deep-learning dqn machine-learning ml neural-network numpy pytorch q-learning reinforcement-learning snake-game supervised-learning supervised-machine-learning terminal

Last synced: 30 days ago
JSON representation

WIP project to use machine learning to play Snake!

• Host: GitHub
• URL: https://github.com/nikorasu/snakeai
• Owner: Nikorasu
• License: gpl-3.0
• Created: 2023-02-05T21:42:23.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2024-04-12T17:41:15.000Z (10 months ago)
• Last Synced: 2024-04-13T21:54:19.937Z (10 months ago)
• Topics: ai, deep-learning, dqn, machine-learning, ml, neural-network, numpy, pytorch, q-learning, reinforcement-learning, snake-game, supervised-learning, supervised-machine-learning, terminal
• Language: Python
• Homepage:
• Size: 64.5 KB
• Stars: 1
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.md

Awesome Lists containing this project

README

        
# Using ML to play Snake

This project is still being worked on, and it's mostly just to satisfy my curiosity.\

Uses a modified version of [MiniSnakes](https://github.com/eliasffyksen/MiniSnakes) and just requires PyTorch and Numpy for now.\

I've setup the base game with rewards, hopefully that can work with for reinforcement learning.\

Also made an old-school bot version (`snake_smrtbot.py` & `snake_tutor.py`) to record data for training.\

The bot version can usually fill over half the play area with snake, highest score I've seen so far was 62/64!\

The current reward system gives 10 points for eating food, -10 for running into itself. As well as a small reward as it moves around, positive values when actions move it closer to food, negative when it's actions head away from food, scaling from 1 up to 9 (spot next to food) as it gets closer.

- `snake.py` is the base game, with modifications for manual or agent to play.

- `agent.py` contains the code for the neural network to train, save, and playback.

- `snake_tutor.py` collects data from tons of best-games played by a simpler bot.

- `snake_user.py` enables manually recording perfect games, including a backspace feature.

- `loading_anim.py` adds a waiting animation to indicate things are processing.

- `snake_smrtbot.py` the bot I came up with, which snake_tutor is based on, not perfect.

- `snake_dumbot.py` an older less effective bot, kept as a reference.

I haven't had much luck getting reinforcement learning working yet, so I've been trying simpler methods.\

Using the _tutor version, I recorded thousands of games to file, along with the action each turn made.\

With that, should be able to train a basic neural network on what move to make for whatever the game state might currently look like.

Example game state:

${\large0~~~0~~~0~~~0~~~0~~~0~~~0~~~0}$\

${\large0~~~0~~~0~~~0~~~{\color{green}1~~~2}~~~0~~~0}$\

${\large0~~~0~~~0~~~0~~~0~~~{\color{green}3}~~~0~~~0}$\

${\large0~~~0~~~0~~~0~~~0~~~{\color{green}4}~~~0~~~0}$\

${\large0~~~0~~~0~~~{\color{green}7~~~6~~~5}~~~0~~~0}$\

${\large0~~~0~~~0~~~{\color{green}8}~~~0~~~0~~~0~~~0}$\

${\large0~~~0~{\color{red}-1}~~0~~~0~~~0~~~0~~~0}$\

${\large0~~~0~~~0~~~0~~~0~~~0~~~0~~~0}$