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
Welcome to the Course!
-Welcome to the second course in the Reinforcement Learning Specialization: Sample-Based Learning Methods, brought to you by the University of Alberta, Onlea, and Coursera. In this pre-course module, you'll be introduced to your instructors, and get a flavour of what the course has in store for you. Make sure to introduce yourself to your classmates in the "Meet and Greet" section!
Monte Carlo Methods for Prediction & Control
-This week you will learn how to estimate value functions and optimal policies, using only sampled experience from the environment. This module represents our first step toward incremental learning methods that learn from the agent’s own interaction with the world, rather than a model of the world. You will learn about on-policy and off-policy methods for prediction and control, using Monte Carlo methods---methods that use sampled returns. You will also be reintroduced to the exploration problem, but more generally in RL, beyond bandits.
Temporal Difference Learning Methods for Prediction
-This week, you will learn about one of the most fundamental concepts in reinforcement learning: temporal difference (TD) learning. TD learning combines some of the features of both Monte Carlo and Dynamic Programming (DP) methods. TD methods are similar to Monte Carlo methods in that they can learn from the agent’s interaction with the world, and do not require knowledge of the model. TD methods are similar to DP methods in that they bootstrap, and thus can learn online---no waiting until the end of an episode. You will see how TD can learn more efficiently than Monte Carlo, due to bootstrapping. For this module, we first focus on TD for prediction, and discuss TD for control in the next module. This week, you will implement TD to estimate the value function for a fixed policy, in a simulated domain.
Temporal Difference Learning Methods for Control
-This week, you will learn about using temporal difference learning for control, as a generalized policy iteration strategy. You will see three different algorithms based on bootstrapping and Bellman equations for control: Sarsa, Q-learning and Expected Sarsa. You will see some of the differences between the methods for on-policy and off-policy control, and that Expected Sarsa is a unified algorithm for both. You will implement Expected Sarsa and Q-learning, on Cliff World.
Planning, Learning & Acting
-Up until now, you might think that learning with and without a model are two distinct, and in some ways, competing strategies: planning with Dynamic Programming verses sample-based learning via TD methods. This week we unify these two strategies with the Dyna architecture. You will learn how to estimate the model from data and then use this model to generate hypothetical experience (a bit like dreaming) to dramatically improve sample efficiency compared to sample-based methods like Q-learning. In addition, you will learn how to design learning systems that are robust to inaccurate models.
completed this course, spending 5 hours a week on it and found the course difficulty to be medium.
To be brief, this is a great course on Reinforcement Learning (RL) and I thoroughly recommend it. This is the second course in the four course Reinforcement Learning specialization from the Alberta Machine Intelligence Institute (AMII) at University of Alberta. The course builds upon the knowledge and skills gained from the first course, (Fundamentals of Reinforcement Learning), and unless you are already very familiar with this field then you should definitely take that course first. AMII is the "home" of Rich Sutton and Andy Barto the authors of Reinforcement Learning an Introduction which i…
To be brief, this is a great course on Reinforcement Learning (RL) and I thoroughly recommend it. This is the second course in the four course Reinforcement Learning specialization from the Alberta Machine Intelligence Institute (AMII) at University of Alberta. The course builds upon the knowledge and skills gained from the first course, (Fundamentals of Reinforcement Learning), and unless you are already very familiar with this field then you should definitely take that course first. AMII is the "home" of Rich Sutton and Andy Barto the authors of Reinforcement Learning an Introduction which is the standard text on RL and is the basis for all the courses in the specialization, (it is available as a free PDF as part of the course material). Sutton & Barto 2018 is also used by Stanford and DeepMind in their RL courses. As with the first course, you get to implement RL algorithms in Jupyter notebooks in Python as weekly programming assignments, so you emerge with practical knowledge at the end of the course. You can check out the syllabus on Coursera.org for details of this course and the other courses in the specialization.
2) Programming assignments are unmanageable - even though you manage to create code that passes unit tests, the grader that attempts to evaluate graphs generated during experiments works as black magic, and if your submission is rejected - you'll never know why
3) regarding the lectures - personally for me they can be replaced by reading relevant chapters of RL book
The specialization is a great way to get into RL. The book is excellent but doing the programming assignments from the book can be a little tedious. My approach was to read the book carefully and do the assignments provided as part of the course. I had a lot of fun.
The course does not, however, require several weeks of effort.
Great course! Strongly based on the textbook by Sutton and Barto, but I think they complement each other greatly. Explanations from the instructors are easy to follow and the online exercises are concrete and illustrative.