Reinforcement Learning beginner to master - AI in Python (Udemy.com)

I liked it because the implementations really help me understand the algorithms. It is a great way to start with the topic. It definitely deserves 5 stars, despite some small issues with the code, e.g., the usage of the old gym libraries. If you cannot use Google Colab, I'd recommend install an older version of Python (I used 3.10) and create a virtual environment to configure and run the notebooks.

The course is good but you just can't rely only on this course if you need in-depth understanding. You need to spend good amount of time in understanding the equations and refer to external sources for derivations, etc. Some parts of this course are not clearly explained especially, the equations and their derivations. If that can be fixed, it can become the best course for "beginner to intermediate RL learning" !

The course covers reinforcement learning concepts well but has several shortcomings. Many Q&A sections are ignored by the instructor, leaving questions unanswered and making it difficult to clarify doubts. Additionally, the course assumes a high level of familiarity with various Python libraries, making it challenging to follow if you don’t have prior experience. I recommend more guidance for beginners, especially in setting up and using these libraries, and more interaction in the Q&A sections for better support.

Very well done course, thank you. Easy to follow and covers what I needed for my personal project. However, I had to go on youtube to understand how the gradient policy theorem was derived (also read section 13.2 of "Reinforcement Learning" second edition book of Sutton and Barto): Everything else is explained but not this important theorem, it would have been nice for this to be explained even as an appendix. Also: few issues with notebooks quite old now and using Python 3.8.

I've purchased a number of reinforcement learning courses on Udemy and this one is easily the best. Although it probably helps that I've learned more about machine learning in the meantime (I did the other courses more than a year ago), there's just something about the presentation of this course that's superior to the others, which I found at the time to be either too much fluff or math intensive to the point of being counterproductive. After doing this course, I was able to read a paper on RL and felt that I understood basically everything in the paper. I'd still recommend having some experience with machine learning and familiarity with basic math/probability notation before diving into the course, but I highly recommend it for those looking to jump from learning about supervised methods to reinforcement learning.

This was a great course to get some basic learning of the field of reinforcement learning! The instructor was so amazing at breaking down the complexities of the algorithms, and I felt like I understood the basics at a high level. The coding exercises were also a great way to translate the algorithms into code. I will feel more confident when I try a bigger problem out.

This is an outstanding first course on RL. One of the very best courses I have taken on Udemy - and I've taken or seen quite a few. The material is well-organized. The production values are excellent. The example code is very well done. This is a challenging course - but if you are serious about building a capability in RL, I would start here.

There's some good content in this course. I did learn quite a bit. But some problems are tough to overlook. I had to take one star off because even though this course is extremely short (11 hours) for what is widely considered among the most challenging topics in machine learning, the length feels padded. Simple things are explained two, three, and sometimes four times. There's a lot of watching the instructor type the same thing repeatedly. E.g., the same structure is used for every neural network, and the videos include it being typed and re-explained every time. It almost seems like some sections were designed to stand alone and thus repeat introductory content. I took off another star for quality/polish issues. There are quite a few unanswered questions in the Q&A section. Some of them are pretty good. Some of the code is out-of-date to the point where it won't work without modification. When the course says to use specific versions of Python libraries from a couple of years ago, believe it. You may also need to roll back to earlier versions of some other libraries (like NumPy) where versions aren't specified. The last two content sections (REINFORCE and A2C) depend on opaque multiprocessing code to parallelize multiple environments. This code has compatibility issues, and it's entirely unnecessary. The computational weight of these tasks comes from the neural networks, not the simple environments in use. Serializing the multiple environments has almost no impact on runtimes. Slides drift between English and Spanish in some spots, including, in a couple of cases, on a single slide. And finally, the audio levels vary significantly between videos, which can be pretty jarring when a very loud one follows a very quiet one. Much (but not all) of the content in this course is taken from "Reinforcement Learning: An Introduction" by Sutton & Barto, which is the de facto reference text in this field. The good news is that the course is cheaper than the book and takes less time. If you want to learn about reinforcement learning, you can do it through this course. Just be prepared for some frustration along the way.

Rationale for my ratings: 1. Pros: a) Large share of course time spent hands-on b) Good overview over different methods for beginners, applied to easily understandable classical problems c) Very tangible, especially through bringing in visualizations of the environment and the learned policies acting on it etc. 2. Cons: a) The technical setup is not very stable and not catering to a broad audience. Building the environment failed on Windows and Ubuntu (both with latest Conda distributions), only worked on Mac. Rendering the environments is not prepared to work on a server, which might be a common setup for everybody using hyperscaler infrastructure for their learning. I needed to approximate the environment with the manual instructions and still debug some dependency issues, taking a lot of time. b) The coding part is good to get hands-on but overly repetitive. I think the author spends too much time doing the same thing over and over again. While this helps make the modules more independent, I think it is not needed for such a small contained course. Copy-pasting from other modules or the solutions provided as well and speeding up the video helps ;-) c) The overview at the end was helpful to get a synopsis of the different methods and their pros and cons. This could have been much more extensive in my point of view.

This is very good course. It helped me to understand overall about RL algorithms. Explanations along with algorithm and related code is really helpful. Initially equations and algorithms parts were being bit difficult to process in mind. Upon further reading the chapters helped to understand more better. After reading all chapters, "looking back" one helps to relate different algorithms at high level. Bit overwhelming at the end, need to understand which algorithm to choose for which type of projects. I think I'll read the book again. The utils and env python scripts explanation may've been better included.