https://github.com/philmod/snake

Toy example of deep reinforcement model playing the game of snake.
https://github.com/philmod/snake

Last synced: about 1 month ago
JSON representation

Toy example of deep reinforcement model playing the game of snake.

• Host: GitHub
• URL: https://github.com/philmod/snake
• Owner: Philmod
• License: mit
• Created: 2016-04-03T21:38:37.000Z (almost 9 years ago)
• Default Branch: master
• Last Pushed: 2016-04-02T17:58:06.000Z (almost 9 years ago)
• Last Synced: 2024-10-28T00:19:10.558Z (3 months ago)
• Language: Python
• Size: 7.9 MB
• Stars: 0
• Watchers: 3
• Forks: 11
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Snake

Toy example of a deep reinforcement model playing the game of Snake. It's a follow up to a similar model that has learned to catch fruit: https://github.com/bitwise-ben/Fruit.



After a few (code) iterations, the model has learned to get to fruit in fewer steps, and stay alive longer.

I'm not sure how much more training is required for the model to become better. The grid size is relatively small (10 x 10), and thus games quickly become difficult as the snake gets longer.

By now, training takes a considerable amount of time, and probably another reinforcement learning algorithm (e.g., A3C) is necessary to keep training time within reasonable bounds.