ITMLPROJ2

Deep q-learning project

Installation

pip install -r requirements.txt

Usage

Each agent file is self contained, so you can run the file which will commence training.

You can set the load boolean to load a trained model, to skip training, if training has already been done.

Agents

• flappy_bird_DQL.py - Deep Q-learning agent. It's DQN, but it is not very good.
• BaselineFlappyQL.py - Q-learning agent. It's our best agent, that we used as a baseline when evaluating other agents.
• AlteredFlappyQL.py - Q-learning agent. It's an agent with higher learning rate, which also performed well, but with higher variance of results.

Bonus

For the bonus competition, we put forth the BaselineFlappyQL agent as the best policy in 1.000.000 frames and AlteredFlappyQL as the fastest learner.