Chevron Left
返回到 Sample-based Learning Methods

學生對 阿尔伯塔大学 提供的 Sample-based Learning Methods 的評價和反饋

4.8
795 個評分
159 條評論

課程概述

In this course, you will learn about several algorithms that can learn near optimal policies based on trial and error interaction with the environment---learning from the agent’s own experience. Learning from actual experience is striking because it requires no prior knowledge of the environment’s dynamics, yet can still attain optimal behavior. We will cover intuitively simple but powerful Monte Carlo methods, and temporal difference learning methods including Q-learning. We will wrap up this course investigating how we can get the best of both worlds: algorithms that can combine model-based planning (similar to dynamic programming) and temporal difference updates to radically accelerate learning. By the end of this course you will be able to: - Understand Temporal-Difference learning and Monte Carlo as two strategies for estimating value functions from sampled experience - Understand the importance of exploration, when using sampled experience rather than dynamic programming sweeps within a model - Understand the connections between Monte Carlo and Dynamic Programming and TD. - Implement and apply the TD algorithm, for estimating value functions - Implement and apply Expected Sarsa and Q-learning (two TD methods for control) - Understand the difference between on-policy and off-policy control - Understand planning with simulated experience (as opposed to classic planning strategies) - Implement a model-based approach to RL, called Dyna, which uses simulated experience - Conduct an empirical study to see the improvements in sample efficiency when using Dyna...

熱門審閱

AA
2020年8月11日

Great course, giving it 5 stars though it deserves both because the assignments have some serious issues that shouldn't actually be a matter. All the other parts are amazing though. Good job

KM
2020年1月9日

Really great resource to follow along the RL Book. IMP Suggestion: Do not skip the reading assignments, they are really helpful and following the videos and assignments becomes easy.

篩選依據:

1 - Sample-based Learning Methods 的 25 個評論(共 156 個)

創建者 JDH

2019年9月22日

Rating 4.3 stars – so far (first two classes combined)

Lectures: 4.0stars

Quizes: 4.0stars

Programming assignments: 4.5stars

Book (Sutton and Barto): 4.5stars

In the spectrum from the theoretical to practical where you have, very roughly,...

(1) “Why”: Why you are doing what you are doing

(2) “What”: What you are doing

(3) “How”: How to implement it (eg programming)…

...this is a “what-how” class.

To cover the “why-what” I strongly recommend augmenting this class with David Silver’s lectures (on Youtube) and notes from a class he gave at UCL. This covers more of the theory/math behind RL but covers less on the coding. Combined together with this class it probably comprises the best RL education you can get *anywhere*, creating a 5-star combo.

http://www0.cs.ucl.ac.uk/staff/d.silver/web/Teaching.html

創建者 Kaiwen Y

2019年10月2日

I spend 1 hour learning the material and coding the assignment while 8 hours trying to debug it so that the grader will not complain. The grader sometimes insists on a particular order of the coding which does not really matter in the real world. Also, grader inconsistently gives 0 marks to a particular part of the problem while give a full mark on other part using the same function. (Like numpy.max) However, the forum is quite helpful and the staff is generally responsive.

創建者 hope

2020年1月25日

This course is ok if you're reading the Sutton & Barto RL book and would like to have some quizzes to follow along. The programming assignments are not really "programming" because you're constrained to type a handful of lines in a few places into a solution that is largely has been prepared for you. With "hints" like "# given the state, select the action using self.choose_action_egreedy(),

# and save current state and action (~2 lines)

### self.past_state = ?

### self.past_action = ?"

it is impossible to get them wrong. These exercises are ok as labs (comparing various algorithms, etc), but the programming part can be done by rote. Coursera has classes with more intense and creative programming assignments and the learning there seems to be much deeper.

創建者 Ivan S F

2019年9月29日

Great course. Clear, concise, practical. Right amount of programming. Right amount of tests of conceptual knowledge. Almost perfect course.

創建者 Manuel B

2019年11月28日

Great course! Really powerful but simply ideas to solve sequential optimization problems based on learning how the environment works.

創建者 Manuel V d S

2019年10月4日

Course was amazing until I reached the final assignment. What a terrible way to grade the notebook part. Also, nobody around in the forums to help... I would still recommend this to anyone interested, unless you have no intention of doing the weekly readings.

創建者 Maxim V

2020年1月12日

Good content, but a lot of annoying issues with grader.

創建者 Andrew G

2019年12月24日

The course needs more support and / or error message output for the programming assignments. Code that seems correct can easily fail the autograder, and the only method of recourse is posting in the forums, which may or may not be received by a moderator.

創建者 Bernard C

2020年3月22日

Course was good but assignment graders were terrible.

創建者 Juan C E

2020年3月7日

Many mistakes with grading and 100% penalty applied for tasks not completed on time, when the rules say that you can submit your assignments and do your quizzes after the deadlines without any penalty.

創建者 Maximiliano B

2020年2月23日

The second course of the specialization is excellent and it provides a solid foundation on sample-based learning methods. The book and the videos complement each other making the learning experience rich and pleasant. The professors explain the content very well and the programming assignments are very interesting to consolidate the knowledge. I had a few issues with the grader and it just returns the score without any message that could help find out what is causing the unexpected behavior. As a suggestion, I would like to suggest that the grader could return any additional information and/or include new unit tests. I am looking forward to begin the next course of the specialization.

創建者 Jonathan

2020年5月9日

Very good class. Has much of the same qualities as its predecessor in the specialization. The methods you learn about though are more exciting, since they go beyond the introductory academic stuff that is not really used in production. I can easily see how Q-Learning, SARSA, and DynaQ architectures are usable in the real world.

Programming assignments are also very similar, but just a **little** bit more challenging. Each assignment has just a touch less handholding than the one before it, although there's still a lot of boilerplate included.

Looking forward to the next class!

創建者 Rishi R

2020年8月3日

There are simply no words to explain how well the instructors have constructed and delivered this. The algorithms were beautifully explained ( unlike in the first course where it was missed) and good intuition was given to the subject. The course is amazing in itself.

Yet if permitted I would like to have an addition. It would be way better if the research papers from which these ideas are introduced are also mentioned, also what other future developments have occurred in that direction if that concept is not visited again.

創建者 Mukund C

2020年3月17日

Excellent Course!! Reading the content and making notes ahead of time before watching the lectures is a MUST!!. The graphics/visuals in describing the concepts in the lectures were very good, especially for a visual learning such as myself. However, I wish there were a few more lectures and the lectures were a little - maybe another 3-5 minutes longer and delved into the derivations/concepts - for example - Bellman Equations to Sarsa/Q-learning/TD.

創建者 Sandesh J

2020年6月8日

The course involves several popular Sample-based RL algorithms with relatable graphical visualizations making it an even smoother transition from the previous course. The lecturers have done a great job of explaining the underlying concepts and highlighting the subtleties of the same. Programming assignments were great which solidified the lessons learned.

創建者 Yover M C C

2020年4月22日

Excelente curso, la calidad de las lecturas y tareas de programación son muy buenas, un curso que no solo te ayuda a mejorar tus habilidades matemáticas y de programación en el tema de aprendizaje por refuerzo, sino también a entender parte del proceso de aprendizaje mediante TD. Un curso que se disfruta mucho! Gracias!.

創建者 Alberto H

2019年10月28日

A great step towards the acquisition of basic and medium complexity RL concepts with a nice balance between theory and practice, similar to the first one.

[Note: the course requires mastering the concepts of the first one in the specialization, so don't start here unless you're sure you master its contents.]

創建者 Parsa V

2020年1月5日

This course is perfect.

You will learn everything about sample-based RL. The programming assignments are harder than the previous course, but you will understand all the algorithms better.

These two courses covered part 1 of the book, and you will build a strong foundation of RL for the future.

創建者 Surya K S

2020年4月12日

One of the more technically challenging courses I've done. Extremely fulfilling, a very good course to go along with the Book - "Reinforcement Learning, An Introduction". The assignments are very engaging, same with the videos, concise and to the point. The forums were incredibly useful.

創建者 Dinh-Son V

2020年7月19日

Excellent course. The amount of information is suited for RL enthusiast. The intuition behind the equation are well introduced. The exercises are challenging, yet interesting. It would have been more enriching to introduce examples not only from the textbook but from other materials.

創建者 Mark J

2019年9月23日

In my opinion, this course strikes a comfortable balance between theory and practice. It is, essentially, a walk-through of the textbook by Sutton and Barto entitled, appropriately enough, 'Reinforcement Learning'. Sutton's appearances in some of the videos are an added treat.

創建者 Damian K

2019年10月5日

Great balance between theory and demonstration of how all techniques works. Exercises are prepared so it is possible to focus on core part of concepts. And if you will you can take deep dive into exercise and how experiments are designed. Very recommended course.

創建者 Yoel S

2020年4月19日

Interesting topic, medium-advanced material, loved the programming assignments (despite technical difficulties with submissions), the textbook is excellent, and the online course is well organized, balanced, and well presented. Thank you to the whole team!

創建者 İbrahim Y

2020年9月29日

The course dives into the methods that are important for basic knowledge of RL intuition. Well designed examples and assignments. It seems somehow easy if a learner knows sth in advance, however, for a new learner of RL, this course is highly recommended.

創建者 Majd W

2019年12月6日

One of the amazing things this specialization stands out in is that it is based on a textbook. if you read from it and watch the lectures, you will have a very good understanding of the material. Also, the programming assignments are very beneficial.