This repository contains two different machine learning models for learning flappy bird.
There are two different Q-Learning classes, one simple with a fixed learning rate and a fixed epsilon. The other class implements learning rate decay and epsilon decay for greater configuration. Surprisingly Q-Learning beats out the Deep-Q-Learning neural network.
The Deep-Q-Learning class is extremely configurable and easy to finetune for different needs. The network architecture is isolated in the network script, and tuning the hyperparameters is done the deep_q_learning script.
For greater detail you can find a pdf report in the repository which explains the project and its results.
Happy Flapping!