What is fast.ai, but for Reinforcement Learning and Deep RL?

I am currently following David Silver’s lectures which are great. I have also been reading Sutton and Barto for a while. I know about Thomas Simonini and Spinning Up by OpenAI.

No course that I know of exists in RL like fast.ai is for DL.

There is no course that closely knits proper theory (I can still hear Jeremy saying, “This is not a version of gradient descent, this is gradient descent”) with hands-on practice and make learners ready to tackle serious problems as fast.ai does. So, what path would you suggest for someone wanting to learn RL, “the fast.ai way”?

So, I am looking for resources, that teach you the fundamentals of RL and also teach you how to get your hands dirty with RL experimentations (with available options such as Spinning Up) and plotting graphs?

NB: I have tried the University of Alberta’s RL Specialization on Coursera (and plan to finish it). But it, to me, seems like a video version of Sutton and Barto’s book w/ some added case studies.

This is the journey that Andrej Karpathy took with RL.

He also recommends the David Silver lectures.

I’m working through Coursera, which I find a good edition to the Barto Sutton book.

How do you rate the David Silver’s and any other resources you’ve tried?

Do you know of any online chat groups, eg slack or IRC?

1 Like

@TomHale Thanks for this resource. Are you sure that that repo is somehow related to Andrej Karpahy?

David Silver’s Lectures are the best holistic treatment to the field you can get. Albeit dense, the lectures provide a high yield for the learners. I thoroughly enjoyed going through them. As a lecture series, I would rate Silver’s lecture series as 9 or 10 on 10. I wouldn’t rate it as a whole learning resource, because it doesn’t teach you to write code for RL. It doesn’t have programming assignments or exercises.

I plan to write a post here outlining my experience with finding a good Deep RL resource soon. When I write it, I will at-mention you. Let me tell you that I have settled on Deep Reinforcement Learning in Action by Alexander Zai from Manning Publication as the most fast.ai-like resource for Deep RL.

I am part of a Discord server which is regular, and has good moderation. Here’s the invite link (good for 5 people for 7 days).

Are you sure that this is the link you wanted to share with me? Please clarify.

I am familiar with this repo as the author shared it with me a few days ago on the fast.ai Discord.