The goal of this project is to parallel train a Reinforement Learning algorithm to play flappy bird. This will be done through techniques discussed in this whitepaper.
- Q learning bot can sucessfully train on the flappy bird env
- Modified game structure to allow multiple games to run simltaneously
- Parallel Combining of Q-Table
- Build Q-Table 'server' that takes in q tables from worker processes and determines the best combinations after every few iterations
- Better Handling of saving q table to file
- Convert global variables into flags(ie: no iterations, no of instances, no gui)
- Optimize code for faster iterations
- Finish documentation
- Clean up folder structure
- Remove unused assets (ie. Audio files)
6 Agents
python flappy.py
Q-learning Tutorial: https://towardsdatascience.com/simple-reinforcement-learning-q-learning-fcddc4b6fe56
The flappy bird game portion of this project was obtained from: https://github.com/sourabhv/FlapPyBird