A collection of Reinforcement Learning algorithms from Sutton and Barto's book and other research papers implemented in Python.
I wrote these notebooks in the winter of March 2017 while I took the COMP 767: Reinforcement Learning [5] class by Prof. Doina Precup at McGill, Montréal in Winter 2017. I highly recommend you to go through the class notes and references of all the papers the intructors have posted on the website.
These notebooks should be used while you read the book and go beyond the same with the referenced papers. I would suggest watching David Silver's videos and reading the book simultaneously. And when you are done with a few chapters, start implementing them. The algorithms follow a pattern and mostly are variants of each other. I have tried my best to explain each notebook's results and possible future directions.
Disclaimer: The code is a little messy. I'd written this when I was not a Pythonista. If you would like to clean them up and want to make it into a nice interface, feel free to contact me. I will be very pleased to collaborate. If you use them then please cite the source and also mention the credits as listed below. Also, email me with ways to improve, let me know if you find any bugs. There might be some lying there, who knows! This repo got over 450 stars on the day I made it public (it is a huge deal to me!). So, now I am motivated to make it much more accessible and easy to understand and use :)
Feel free to reach me at pulkit.khandelwal@mail.mcgill.ca or see my website here
Special Credits:
[1] Denny Britz
[2] Monica Patel
[3] Sutton and Barto
[4] David Silver