This is my rifle. There are many like it, but this one is mine.

- Major General William H. Rupertus

A generic utility library for deep reinforcement learning. I built this throughout my PhD as sort of a personal swiss-army knife for my research. There are other libraries that do similar things, this one is mine.

Highlights include:

• Implementations of several deep reinforcement learning algorithms, including Proximal Policy Optimization, Augmented Random Search, and Soft Actor Critic. All using a unified interface.

• Neural network utilities, Including supervised learning functionality mimicking keras' model.fit in pytorch, and an interface to make various MLPs, RBFs etc that are compatible with the rest of this library.

• Tests Known good hyper parameters across different environements for the algorithms.

• Custom OpenAI gym environments, That I've studied at some point or another in my research.

• Experiment utilities, For saving RL experiments along with meta data, logs, and a mechanism to restore the trained agents.

If you just want to get your feet wet with deep RL there are better places to start, I usually recomend stable-baselines3. Most of the value of something like this is that I personally know it inside and out, but that won't be an advantage to someone out.

If you are a student working with me, or someone looking to extend my previous work, you've come to the right place.

This is a python package, so you can install it with:

pip install git+https://github.com/sgillen/seagul

Or if you want to make changes:

git clone https://github.com/sgillen/seagul
cd seagul
pip install -e .

I think that the tests folder linked above is a good starting point, and many files have a main restricted portion that shows some basic usage.