Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
A Long Peek into Reinforcement Learning (lilianweng.github.io)
147 points by shry4ns on Oct 15, 2018 | hide | past | favorite | 7 comments


I'm currently studying this topic (preparing an online course about it actually) and here are the best resources I found so far:

Full Courses

- From David Silver: http://www0.cs.ucl.ac.uk/staff/d.silver/web/Teaching.html

- From Yandex: https://yandexdataschool.com/edu-process/rl

- From Sergey Levine: http://rail.eecs.berkeley.edu/deeprlcoursesp17/index.html

Articles:

- Q-learning on Taxi-v2, very good basic explanation: https://www.learndatasci.com/tutorials/reinforcement-q-learn...

- Q-learning and DQN, goes a bit further: https://neuro.cs.ut.ee/demystifying-deep-reinforcement-learn...

- "Pong from Pixels from Karpathy", introduces DQN and PG: https://karpathy.github.io/2016/05/31/rl/

Baseline implementations:

- "RL-Adventures", super clean Pytorch implementations: https://github.com/higgsfield/RL-Adventure (DQN) and https://github.com/higgsfield/RL-Adventure-2 (PG)

- Repo for the Deep Reinforcement Learning Nanodegree: https://github.com/udacity/deep-reinforcement-learning

- stable-baselines, a better documented fork of OpenAI baselines: https://github.com/hill-a/stable-baselines

If you have more high-quality resources please share! :D


I'm a TA for a RL class and would love to hear more about your online course. If you have any content that's ready for viewing, that would be great.


Cute, but insufficient to understand unless you already understand the concepts and deep mathematics involved.

It doesn't tell you why these methods work at learning so many problems. It only tells you how they work which is not enough and will bite you the moment you fall into a pitfall. (and there are many more than mentioned)

And most importantly, it doesn't tell you how to construct working fitness/target functions or why some generic ones work or don't, and when.


>In this post, we are gonna briefly go over the field of Reinforcement Learning (RL), from fundamental concepts to classic algorithms. Hopefully, this review is helpful enough so that newbies would not get lost in specialized terms and jargons while starting.

Read the mission statement at the very top. This isn't supposed to be Sutton and Barto.


"This isn't supposed to be Sutton and Barto"

It reads quite a lot like a synopsis of Section I of Sutton and Barto, down to the ordering of the section headings. Not that the writing isn't clear and concise, but it reads more like a cheat sheet than like serious exposition.

Sutton and Barto available below [0] for comparison. The google drive link comes from Sutton's website [1] so I think it's kosher from a copyright perspective.

[0] https://drive.google.com/file/d/1opPSz5AZ_kVa1uWOdOiveNiBFiE...

[1] http://incompleteideas.net/book/the-book-2nd.html


Cute, but insufficient to understand unless you already understand the concepts and deep mathematics involved.

I don't get this comment. Does this article claim to explain all the "concepts and deep mathematics"? From what I can see, the answer is no:

"Hopefully, this review is helpful enough so that newbies would not get lost in specialized terms and jargons while starting"

This seems to be objecting that the site isn't something that it never claimed to be. Sometimes you need a guide for newbies that doesn't go into all the details. Not every single resource on a given topic needs to be a complete reference manual with every detail that exists on the field. There is room for a variety of resources, at various levels of detail, aimed at a variety of different audiences, no?





Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: