A Concise introduction to key ideas in RL.
pip install fastrl
The following notes is targeted at people interested in Reinforcement Learning looking to get upto speed without sacfricing on rigor.The necessary math is tackled but with computation first approach.Basic ability to write and understand simple programs and high school level math is all that is required.For readers looking to understand the latest developments in this field refer amazing Spinning Up RL
While reading the text reader should constantly be asking two questions :
How can I implement this ? :
Brings practical computational problems into the picture and ensures actual understanding.
How can I be sure what I'm doing will work(If implemented correctly) ? :
As we will see RL needs lot of tricks and hacks to make it work. Whenever some notation and math help us in choosing sensible things,we should do it. This is where the rigor part of this notes comes into picture.This will be essential in removing the magic out of many **SOTA** algorithms in RL.