Chevron Left
Back to Prediction and Control with Function Approximation

Learner Reviews & Feedback for Prediction and Control with Function Approximation by University of Alberta

690 ratings
126 reviews

About the Course

In this course, you will learn how to solve problems with large, high-dimensional, and potentially infinite state spaces. You will see that estimating value functions can be cast as a supervised learning problem---function approximation---allowing you to build agents that carefully balance generalization and discrimination in order to maximize reward. We will begin this journey by investigating how our policy evaluation or prediction methods like Monte Carlo and TD can be extended to the function approximation setting. You will learn about feature construction techniques for RL, and representation learning via neural networks and backprop. We conclude this course with a deep-dive into policy gradient methods; a way to learn policies directly without learning a value function. In this course you will solve two continuous-state control tasks and investigate the benefits of policy gradient methods in a continuous-action environment. Prerequisites: This course strongly builds on the fundamentals of Courses 1 and 2, and learners should have completed these before starting this course. Learners should also be comfortable with probabilities & expectations, basic linear algebra, basic calculus, Python 3.0 (at least 1 year), and implementing algorithms from pseudocode. By the end of this course, you will be able to: -Understand how to use supervised learning approaches to approximate value functions -Understand objectives for prediction (value estimation) under function approximation -Implement TD with function approximation (state aggregation), on an environment with an infinite state space (continuous state space) -Understand fixed basis and neural network approaches to feature construction -Implement TD with neural network function approximation in a continuous state environment -Understand new difficulties in exploration when moving to function approximation -Contrast discounted problem formulations for control versus an average reward problem formulation -Implement expected Sarsa and Q-learning with function approximation on a continuous state control task -Understand objectives for directly estimating policies (policy gradient objectives) -Implement a policy gradient method (called Actor-Critic) on a discrete state environment...

Top reviews

Dec 1, 2019

Well peaced and thoughtfully explained course. Highly recommended for anyone willing to set solid grounding in Reinforcement Learning. Thank you Coursera and Univ. of Alberta for the masterclass.

Jun 24, 2020

Surely a level-up from the previous courses. This course adds to and extends what has been learned in courses 1 & 2 to a greater sphere of real-world problems. Great job Prof. Adam and Martha!

Filter by:

76 - 100 of 126 Reviews for Prediction and Control with Function Approximation

By Stewart A

Oct 31, 2019

Simply the best course on this topic.

By Junchao

May 29, 2020

Very good and self-oriented course!

By Fernando A S G

Mar 26, 2021

Excellent course! Thanks a lot!

By Wei J

Oct 11, 2020

It is a very perfect RL course.

By Antonis S

May 30, 2020

Really a well-prepared course!

By Ignacio O

Nov 29, 2019

Really good, I learned a lot.


May 2, 2020

Great speakers and content!

By Majd W

Feb 1, 2020

Very practical course.

By 李谨杰

Jun 17, 2020

Excellent class !!!

By Mohamed A

Sep 11, 2021

v​ery good course

By Hugo T K

Aug 18, 2020

Excellent course.

By Murtaza K B

Apr 25, 2020

Excellent course

By Ivan M

Aug 30, 2020

Just brilliant

By Oriol A L

Nov 19, 2020

Very good!

By Cheuk L Y

Jul 8, 2020

Very good!

By Jialong F

Feb 23, 2021


By Justin O

May 18, 2021



Feb 27, 2021


By Ananthapadmanaban, J

Jul 19, 2020

I am disappointed with policy gradients being introduced on last week of the 3rd course. The instructors need to understand that 12 weeks is too much for introduction before starting a good project to implement the concepts with a hope to better understand them (course 4). Policy gradients should have been introduced in week 3/4 of course 2 itself. The content before that should be made more efficient (4 weeks to understand until q-learning/sarsa and 2 weeks to understand function approximation should be enough). I realized after course 2 that Andrew Ng has 3/4 videos on RL in the recently released ML class from Stanford. I am yet to go through them, but I feel they may explain these faster with same amount of rigour. However, the stanford class assignments are not public, which makes this course still useful because of the assignments. However, thanks to the instructors for this course.


Jun 18, 2021

T​his is a good course, but I continue to be disappointed in the lack of detail in the lectures. I fill in the detail with the Deep Mind lectures on Reinforcement Learning by David Silver. The programming assignments are difficult, not because they are challenging, but because the data structures are not well explained and the conceptual connections between the equations in the book and the code structures used for the implementation are not clear. It's like being given somebody's not-very-well-documented code and trying to figure out what they were thinking. All that said, I think that the course offers a lot and I have learned a lot from it so far.

By Luiz C

Oct 3, 2019

Almost perfect, except two ~minor objections:

1/ the learning content between the 4 weeks is quite unbalanced. The initial weeks of the course are well sized, whereas week #3 and week #4 feel a touch light. It feels like the Instructors rushed to make the Course available online, and didn't have time to put as much content as they wished in the last weeks of the Course

2/ there are too many typos in some notebooks (specifically notebook of week #3). It gives the impression it was made in a rush, and nobody read over it again. Besides there seems to currently be some issue with this assignment

By Luka K

Jan 4, 2021

It is a good introduction to prediction and control with function approximation. Combining book and instructros results in a simple and nice explanation. What keeps it from the perfect grade are the examples. It would be nice if there are more examples and explained in a more detailed way why and how the example works. For example sometimes instructors would just say that the robot can use this, and that is mostly it. The other thing is more interactive project work. For example I would like to see how is my pendulum moving after N number of episodes. I would feel more satisfactory then.

By Dmitry S

Jan 5, 2020

Definitely a course to take to learn the ropes of RL. For this course, it is critical to follow and math. I'd love to give 5 stars to this course but will however take one away since the course could benefit a lot if the math was made a bit simpler to follow. The book referenced in the course is excellent and does help, but still, some more pedagogical repetition/rephrase, simplification of notation, a bit slower pace of narration would make the course even better. Having said that, this seems to be the best course available at this time. Many thanks to tutors.

By Hadrien H

Feb 4, 2021

I really appreciate that this course gives more hands on and assignments exercises. Really helped a lot in the understanding of the theory. As the books gets deeper into concepts and complexity so does the class, which is nice, but I felt like the depth and complexity in which the online class goes does not really keep up with the book content. Not only by skipping chapters but also by staying a bit at a too high level sometimes. Still a very good course again and really accessible, entertaining and resourceful material and instructors.

By Steven W

May 11, 2021

It's a great course, and they cover the basics of function approximation. The instructors were clear and knowledgeable, and the content that was covered was solid.

However, they skip some content that I feel is really important for modern RL, specifically the "deadly triad" regarding the convergence of off-policy approximate TD methods. They also don't discuss or link to papers on PPO or other recent advancements in RL, and I was hoping to learn more about those in particular.