Ensemble Reinforcement Learning

This repositeroy presents the code for a three-part series on implementing Ensemble Learning in Cartpole using Q-learning, Deep SARSA, and Deep REINFORCE.

Briefly, ensemble learning is a method of combining multiple learning models to produce a single, more robust learner. For the purposes of this repository, we employed a simple environment and focus on understanding the learning models and the ensemble logic.

Getting Started

Understanding the Code

This repo includes four main jupyter notebooks:

1. A Deep SARSA implementation.
2. A Q-learning implementation.
3. A Deep REINFORCE implementation. See the trained agent here.
4. An Ensemble learning implementation using the three agents above.

Tutorials

A series of tutorials, explaining this code is available on medium.

0. An introductory tutorial.
1. A tutorial on Deep SARSA.
2. A tutorial on Q-learning.
3. A tutorial on Deep REINFORCE.
4. A tutorial on Ensemble learning.

Prerequisites

To run the above code, make sure to install -

• Keras
• Tensorflow
• OpenAI Gym

Authors

@gelanat, @JoeSnow7, @AshNguyen, and @karnigili.