This repository includes Python implementations of Sutton's & Barto's Reinforcement Learning Book (Ed. 2) algorithms. The latest draft of the book's second version can be found in this link. In this repository you may also find the solutions of the book exercises that I've been implementing.
This repository is divided into several folders, each for a chapter of the book. In general, every episode is related to a set of Reinforcement Learning algorithms (e.g. Time Difference Methods).
Additionally, in all chapter-directories, a notebooks sub-folder is present. Such files are interactive Jupyter Notebooks, where users can interact with OpenAI environments and the algorithms available.
The following Python files, found in the home directory, are the most relevant for this project:
classes.py
Contains the implementation of common models used in Reinforcement Learning tasks. An example of this are Policy classes (e.g. e-greedy), Action Value Functions, and relevant data structures (e.g. N-Step Memory).
utils.py
Contains auxiliary methods such as printing and displaying environment interaction logs, which are used by other modules.
visualize.py
Contains methods that enable the visualization of statistics calculated from the experience of agents that have ran in environments.
In your terminal, run:
git clone git@github.com:diegoalejogm/Reinforcement-Learning.git
cd Reinforcement-Learning
Then, install all dependencies with PIP using:
pip install -r requirements.txt
And run a jupyter process in order to visualize and interact with all the available Jupyter notebooks:
jupyter notebook .
This section is for myself, but feel free to fork the repo and add your contributions!
- Add Tests to Data Structures and Models.
MIT