In this final course, you will put together your knowledge from Courses 1, 2 and 3 to implement a complete RL solution to a problem. This capstone will let you see how each component---problem formulation, algorithm selection, parameter selection and representation design---fits together into a complete solution, and how to make appropriate choices when deploying RL in the real world. This project will require you to implement both the environment to stimulate your problem, and a control agent with Neural Network function approximation. In addition, you will conduct a scientific study of your learning system to develop your ability to assess the robustness of RL agents. To use RL in the real world, it is critical to (a) appropriately formalize the problem as an MDP, (b) select appropriate algorithms, (c ) identify what choices in your implementation will have large impacts on performance and (d) validate the expected behaviour of your algorithms. This capstone is valuable for anyone who is planning on using RL to solve real problems.
To be successful in this course, you will need to have completed Courses 1, 2, and 3 of this Specialization or the equivalent.
By the end of this course, you will be able to:
Complete an RL solution to a problem, starting from problem formulation, appropriate algorithm selection and implementation and empirical study into the effectiveness of the solution.
A Complete Reinforcement Learning System (Capstone)
University of Alberta and Alberta Machine Intelligence Institute via Coursera
Overview
Syllabus
- Welcome to the Final Capstone Course!
- Welcome to the final capstone course of the Reinforcement Learning Specialization!!
- Milestone 1: Formalize Word Problem as MDP
- This week you will read a description of a problem, and translate it into an MDP. You will complete skeleton code for this environment, to obtain a complete MDP for use in this capstone project.
- Milestone 2: Choosing The Right Algorithm
- This week you will select from three algorithms, to learn a policy for the environment. You will reflect on and discuss the appropriateness of each algorithm for this environment.
- Milestone 3: Identify Key Performance Parameters
- This week you will identify key parameters that affect the performance of your agent. The goal is to understand the space of options, to later enable you to choose which parameter you will investigate in-depth for your agent.
- Milestone 4: Implement Your Agent
- This week, you will implement your agent using Expected Sarsa or Q-learning with RMSProp and Neural Networks. To use NNs, you will have to use a more careful stepsize selection strategy, which is why you will use RMSProp. You will also verify the correctness of your agent.
- Milestone 5: Submit Your Parameter Study!
- This week you will identify a parameter to study, for your agent. Once you select the parameter to study, we will provide you with a range of values and specific values for other parameters. You will write a script to run your agent and environment on the set of parameters, to determine performance across these parameters. You will gain insight into the impact of parameters on agent performance. You will also get to visualize the agents that you learn. Your parameter study will consist of an array of values that we will check for correctness.
Taught by
Martha White and Adam White
Related Courses
Related articles
Reviews
4.7 rating, based on 26 Class Central reviews
4.7 rating at Coursera based on 630 ratings
-
It was a remarkable experience; this course provided me with insights into how various materials covered in previous courses seamlessly came together.
-
This course really helps you see Reinforcement Learning from the problem definition through the system build up to the solution. It has well written exercises and videos that help you understand the core and basics of the problem solving with RL. I only can recommend it.
-
I really enjoyed the experience I got through completing this course. The assignments were well designed.
-
Steven WeissOverall a good class that covers the more practical aspects of RL, though I think branding it as a "capstone" is a little generous. The class covers important things like hyperparameter tuning and reward function creation and tuning. Much of the co… -
I enjoyed every bit of it. Learned the fundamentals of RL from tabular down to function approximation.
Implemented the deep learning models from scratch.
Really a good course for understanding the fundamentals of RL -
Absolutely Fantastic Course!! Puts everything together nicely. Loved the Capstone project. I wish the Capstone project was a little bit longer when doing some exploration of hyper-parameter tuning. Both the instructors are beyond wonderful. They really know their subject matter and explain it really well. Thanks to them, their team and the institution to put this Specialization together. I wish there was a little bit more walk-through of some proofs that are there in the textbook to get a better theoretical grounding - but not a complaint - just an observation/suggestion. If there is a way to incorporate reading/implementing a paper - that would be great as well!! Again, just a suggestion.
-
Really comprehensive and good course that follows the book by Richard Sutton, but everything which seems to academical and formal in the book, in course is better explained with examples. I think specialisation provides very good foundations with really interesting programming exercises which are carefully designed. In some moments, I have felt overwhelmed with new concepts while old ones are no recalled, but at the end, very good recapitulation was done. Course demands from student to be patient and proceed, but also to fully understand basic concepts before getting new ones.
-
The idea of dedicating a whole course to a practical project was indeed very good. However I think that this idea was not exploited as deeply as it could have been. The project itself is actually a notebook just a little longer than usual. I would have left more to do to students, and maybe they could have used lectures to give more explanations and hints for the practical part, or to do some programming together. Anyway, the course is overall good, and it also introduces some new concepts, like experience replay.
-
This was my least-favorite of the RL Specialization on Coursera. Most of the lectures were repeats, and the programming assignments were either way too easy (#1 and #3) or way too involved/difficult (#2).
However, I did love the idea of pulling all the knowledge together into a single course with a unified project. The "Meeting with..." videos and the guest lectures from this class were definitely some of my favorite videos/lectures out of the entire specialization. I just wish there were more of them! -
A nice starting point to start learning about the foundation of Reinforcement Learning. It does not go into "fancy deep RL" algorithms but it gives a solid foundation. Personally, reading Rich Sutton's RL book is very abstract. At first, it is enjoyable, but the more you deep dive into the book, without clear context, it becomes very abstract. This course guides me to finish the book with a solid grasp of what's being told on the book.
Highly recommend it if you are starting in RL. -
Nice course, it contains some insights I wish I saw in the first course of the specialization.
It has the same defaults as the other course of the specialization: the notebooks have to fit exactly the ones created by the team, it lacks some creativity in the approaches, by asking us to fill the holes in algorithms, but it's nice to see the whole implementation behind an RL algorithm in practice.
I recommend the whole specialization as an introduction to RL. -
The course was excellent and the syllabus was designed really well such that it covers everything from basic. The instructors- Adam and Martha are just amazing in explaining each concept in a crystal clear manner. Only thing was the capstone project could have been more difficult and if possible included some real world application rather than some simulation otherwise this is the best course for someone who wants to begin their journey in RL.
-
A nice way to end this great specialization! The project choice of the lunar module is exciting and everything that was covered in the previous courses becomes clear in the context. The two instructors are very clear and involved, making the learning process pleasant. I just wished the project was divided into more pieces and more work for the student instead of the three assigments with very different difficulty levels.
-
This course was a great finale to the Reinforcement Learning Specialization. The coding segments were challenging but completable, and the lectures were clear and informative. I really liked how they included several mini-lectures by professionals in the field of machine learning. They were not required to complete the course, but they were super insightful.
-
I have completed this course, together with the other three courses are Coursera for the RL specialization. The course is overally very well designed, with comprehensive coverage from various theories, practical considerations and sample codes demonstrations. Highly recommend this course as your starting point of the RL journey!
-
For those looking for Reinforcement Learning courses I highly recommend this one. If you are starting, I would recommend to learn first Probability and Statistics (usually initial chapters of books, not everything), then go to a Artificial Neural Networks course and then go to this one, which uses them all!
-
Nice capstone to finish a very great specialisation! Programming assignments might be a bit more challenging from my perspective. But in that case they could be overly hard for someone with less experience. Overall nicely balanced course. Many thanks to the authors!
-
I really loved the class format: reading passage of the reference book and then follow videos to strengthen the learned theory. The capstone project is really fun and well guided. I highly recommend this specialization, a big thank you to all the instructors !
-
Just the right course to sum up all the concepts learned so far in RL specialization by University of Alberta. The project is challenging and the descriptions are very neat and to the point. It really helps us test our understanding of the RL concepts.
-
Excellent summary of the entire specialization and good capstone project. Could be better if students were allowed to tweak reward parameters and see how it affects agent performance
