This summer (2020), I have decided to take CS 285 (http://rail.eecs.berkeley.edu/deeprlcourse/) a course on Deep Reinforcement Learning.
The link above provides references to course lecture videos, slides, and homework assignments. My goal is to get the most out of this series, and unfortunately going directly into the homework assignments will not do that for two reasons.
- It's in tensorflow :-)
- The environments they train in sometimes are MuJoCo, and I don't have the time to figure out that installation process.
So instead, I have decided to take these homework assignments and lessons, and combine them with the numerous tutorials available online, so they can be delivered in a more accessible way avoiding the pitfalls listed above.
Further documentation provided in the corresponding sub-directories, and references will be provided accordingly.
I want to also add I've drawn a lot of these algorithm implementations from a few different sources, mainly
https://github.com/nikhilbarhate99/PPO-PyTorch
https://pytorch.org/tutorials/intermediate/reinforcement_q_learning.html
and
https://www.youtube.com/channel/UC58v9cLitc8VaCjrcKyAbrw
This repository is mostly about getting those methods to work on different environments and modifying small parts of those scripts for improved logging, performance, or understanding.