heojay/ReinforcementLearningSnake

4-grade

ReinforcementLearningSnake

Research project for 1DL073 Natural Computation Methods for Machine Learning at Uppsala University.

Supervisor

• Chencheng Liang

Researchers

• Johan Alfredéen
• Jae Won Heo
• Adam Ross

Blind, OpenAI Gym

This implementation uses a modified version of the OpenAI gym from https://github.com/grantsrb/Gym-Snake

• Q-learning
• Epsilon decay

Instructions

To train

python3 rl_snake/rl_snake_trainer.py -t

To replay

python3 rl_snake/rl_snake_trainer.py -r

Blind, custom

Man cave developed Snake game and reinforcement learning environment, not based on anything prior aside from the concept of the classic Snake game. Trains and replays the following algorithms:

• Q-learning
• SARSA

Instructions

To train Q-learning

python3 rl_algorithms.py -q -l

To train SARSA

python3 rl_algorithms.py -s -l

To playback Q-learning training

python3 rl_algorithms.py -q -o

To playback SARSA training

python3 rl_algorithms.py -s -o

To play the Snake game

python3 rl_algorithms.py -m 

Semi-blind, PyGame Snake

• Q-learning
• Distance-based learning