Search icon CANCEL
Subscription
0
Cart icon
Your Cart (0 item)
Close icon
You have no products in your basket yet
Save more on your purchases now! discount-offer-chevron-icon
Savings automatically calculated. No voucher code required.
Arrow left icon
Explore Products
Best Sellers
New Releases
Books
Videos
Audiobooks
Learning Hub
Conferences
Free Learning
Arrow right icon
Arrow up icon
GO TO TOP
PyTorch 1.x Reinforcement Learning Cookbook

You're reading from   PyTorch 1.x Reinforcement Learning Cookbook Over 60 recipes to design, develop, and deploy self-learning AI models using Python

Arrow left icon
Product type Paperback
Published in Oct 2019
Publisher Packt
ISBN-13 9781838551964
Length 340 pages
Edition 1st Edition
Languages
Tools
Arrow right icon
Author (1):
Arrow left icon
Yuxi (Hayden) Liu Yuxi (Hayden) Liu
Author Profile Icon Yuxi (Hayden) Liu
Yuxi (Hayden) Liu
Arrow right icon
View More author details
Toc

Table of Contents (11) Chapters Close

Preface 1. Getting Started with Reinforcement Learning and PyTorch FREE CHAPTER 2. Markov Decision Processes and Dynamic Programming 3. Monte Carlo Methods for Making Numerical Estimations 4. Temporal Difference and Q-Learning 5. Solving Multi-armed Bandit Problems 6. Scaling Up Learning with Function Approximation 7. Deep Q-Networks in Action 8. Implementing Policy Gradients and Policy Optimization 9. Capstone Project – Playing Flappy Bird with DQN 10. Other Books You May Enjoy

Performing off-policy Monte Carlo control

Another MC-based approach to solve an MDP is with off-policy control, which we will discuss in this recipe.

The off-policy method optimizes the target policy, π, using data generated by another policy, called the behavior policy, b. The target policy performs exploitation all the time while the behavior policy is for exploration purposes. This means that the target policy is greedy with respect to its current Q-function, and the behavior policy generates behavior so that the target policy has data to learn from. The behavior policy can be anything as long as all actions in all states can be chosen with non-zero probabilities, which guarantees that the behavior policy can explore all possibilities.

Since we are dealing with two different policies in the off-policy method, we can only use the common steps in episodes that take place...

lock icon The rest of the chapter is locked
Register for a free Packt account to unlock a world of extra content!
A free Packt account unlocks extra newsletters, articles, discounted offers, and much more. Start advancing your knowledge today.
Unlock this book and the full library FREE for 7 days
Get unlimited access to 7000+ expert-authored eBooks and videos courses covering every tech area you can think of
Renews at €18.99/month. Cancel anytime