RLib

The library that I built while learning some Reinforcement Learning algorithms. The library is built on top of the Gymnasium and PyTorch frameworks. It is designed to be as modular as possible. The documentation, built using Sphinx can be found here.

Installation

Code

To install the package simply run the following command:

pip install .

Remember that, if using a virtual environment, you must activate it before running the command. Furthermore, if using conda, you must call conda install pip before installing the package. Otherwise, the package will be installed using another pip that is not the one from the conda environment. To run the library on all gymnasium environments, other dependencies are required. To install them, run the following command:

pip install gymnasium[all]

or

pip install "gymnasium[all]"

Note that installing Box2D and MuJoCo is not trivial and the installation is quite dependent on the OS. For example, to install MuJoCo on Windows, we add to install MuJoCo version 1.50.1 while the latest version when we write those lines is 2.1.0.

Documentation

To generate the documentation, one must install Sphinx and the Furo theme:

pip install sphinx
pip install furo

Then, to generate the documentation, run the following command:

cd docs
make html

The documentation will be generated in the docs/build/html folder. To open it, simply open the index.html in your browser.

Examples

Here are a few examples of trained agents using the library, the code used to produce those results can be found in the scripts folder:

Q-Learning on Mountain Car	Deep QLearning on Lunar Lander	Evolution Strategy on Flappy Bird	Deep Deterministic Policy Gradient on Half Cheetah	PPO on Bipedal Walker