You're reading from Deep Reinforcement Learning Hands-On Apply modern RL methods to practical problems of chatbots, robotics, discrete optimization, web automation, and more

Product type Paperback

Published in Jan 2020

Publisher Packt

ISBN-13 9781838826994

Length 826 pages

Edition 2nd Edition

Languages

Python

Tools

Deep Reinforcement Learning

Concepts

Chatbots

Author (1):

Maxim Lapan

View More author details

Table of Contents (28) Chapters

Preface

1. What Is Reinforcement Learning?

2. OpenAI Gym FREE CHAPTER

3. Deep Learning with PyTorch

4. The Cross-Entropy Method

5. Tabular Learning and the Bellman Equation

6. Deep Q-Networks

7. Higher-Level RL Libraries

8. DQN Extensions

9. Ways to Speed up RL

10. Stocks Trading Using RL

11. Policy Gradients – an Alternative

12. The Actor-Critic Method

13. Asynchronous Advantage Actor-Critic

14. Training Chatbots with RL

15. The TextWorld Environment

16. Web Navigation

17. Continuous Action Space

18. RL in Robotics

19. Trust Regions – PPO, TRPO, ACKTR, and SAC

20. Black-Box Optimization in RL

21. Advanced Exploration

22. Beyond Model-Free – Imagination

23. AlphaGo Zero

24. RL in Discrete Optimization

25. Multi-agent RL

26. Other Books You May Enjoy

27. Index

Double DQN

The next fruitful idea on how to improve a basic DQN came from DeepMind researchers in the paper titled Deep Reinforcement Learning with Double Q-Learning ([3] van Hasselt, Guez, and Silver, 2015). In the paper, the authors demonstrated that the basic DQN tends to overestimate values for Q, which may be harmful to training performance and sometimes can lead to suboptimal policies. The root cause of this is the max operation in the Bellman equation, but the strict proof is too complicated to write down here. As a solution to this problem, the authors proposed modifying the Bellman update a bit.

In the basic DQN, our target value for Q looked like this:

Q'(s_t₊₁, a) was Q-values calculated using our target network, so we update with the trained network every n steps. The authors of the paper proposed choosing actions for the next state using the trained network, but taking values of Q from the target network. So, the new expression for target Q-values will look...