Use Deep Q network to solve maze problem generated randomly
IMPORTANT: comment the env.render() can obtain results quicker (rendering surely much slower than CPU staffssssss).
gym_maze: the gym library for generating experiment environment
How to run? python maze.py
Why it's a mess? I changed the value of a lot of parameters to do experiments.
Any innovation point? See the last part in the report.
the freq=xxx folders? Stored the experiment result.
I think you can find everything you need in my report.
For Q-learning version maze solver, see:https://github.com/saaries/Maze_reinforcement_learning
You should get logs like this if the program works well:
