Implementation of the Flappy Bird game with pygame and reinforcement learning AI bots: greedy learner and Q-learner that learns the skills to play the game.
This work was done for a university assignment and it is based on flappybird-qlearning-bot and FlappyBirdBotsAI.
To run the code you'll need the following packages:
- pygame
- numpy
- sklearn
- matplotlib
- pickle
- collections
- random
Normally the only package that you'll need to install additionally is pygame, which you can do by
pip install pygame
flappy.py-Main file, runs the game with visual display, either you or the AI agent can play the game
greedy_agent.py-Greedy agent class.
q_learning_agent.py-Q-learner agent class.
utils.py-Utility script to 1) plot the earned scores in function of training iterations; 2) to run multiple concurrent subprocesses (not parallel) with python module to speed up training of the reinforcement learning agent.
conc_run.sh- Bash script which executes multiple instances of 'flappy.py' which can run parallel on multicore devices. This helps to speed up training of the reinforcement learning agent.
Screenshot of concurrent training:
sprites and hitmasks for the visual output
qvalues: state and action pairs for the Q-learner agent
q_learning_scores, greedy_scores achieved score after each play