What this book covers

Chapter 1, Fundamentals of Reinforcement Learning, helps you build a strong foundation on RL concepts. We will learn about the key elements of RL, the Markov decision process, and several important fundamental concepts such as action spaces, policies, episodes, the value function, and the Q function. At the end of the chapter, we will learn about some of the interesting applications of RL and we will also look into the key terms and terminologies frequently used in RL.

Chapter 2, A Guide to the Gym Toolkit, provides a complete guide to OpenAI's Gym toolkit. We will understand several interesting environments provided by Gym in detail by implementing them. We will begin our hands-on RL journey from this chapter by implementing several fundamental RL concepts using Gym.

Chapter 3, The Bellman Equation and Dynamic Programming, will help us understand the Bellman equation in detail with extensive math. Next, we will learn two interesting classic RL algorithms called the value and policy iteration methods, which we can use to find the optimal policy. We will also see how to implement value and policy iteration methods for solving the Frozen Lake problem.

Chapter 4, Monte Carlo Methods, explains the model-free method, Monte Carlo. We will learn what prediction and control tasks are, and then we will look into Monte Carlo prediction and Monte Carlo control methods in detail. Next, we will implement the Monte Carlo method to solve the blackjack game using the Gym toolkit.

Chapter 5, Understanding Temporal Difference Learning, deals with one of the most popular and widely used model-free methods called Temporal Difference (TD) learning. First, we will learn how the TD prediction method works in detail, and then we will explore the on-policy TD control method called SARSA and the off-policy TD control method called Q learning in detail. We will also implement TD control methods to solve the Frozen Lake problem using Gym.

Chapter 6, Case Study – The MAB Problem, explains one of the classic problems in RL called the multi-armed bandit (MAB) problem. We will start the chapter by understanding what the MAB problem is and then we will learn about several exploration strategies such as epsilon-greedy, softmax exploration, upper confidence bound, and Thompson sampling methods for solving the MAB problem in detail.

Chapter 7, Deep Learning Foundations, helps us to build a strong foundation on deep learning. We will start the chapter by understanding how artificial neural networks work. Then we will learn several interesting deep learning algorithms, such as recurrent neural networks, LSTM networks, convolutional neural networks, and generative adversarial networks.

Chapter 8, A Primer on TensorFlow, deals with one of the most popular deep learning libraries called TensorFlow. We will understand how to use TensorFlow by implementing a neural network to recognize handwritten digits. Next, we will learn to perform several math operations using TensorFlow. Later, we will learn about TensorFlow 2.0 and see how it differs from the previous TensorFlow versions.

Chapter 9, Deep Q Network and Its Variants, enables us to kick-start our deep RL journey. We will learn about one of the most popular deep RL algorithms called the Deep Q Network (DQN). We will understand how DQN works step by step along with the extensive math. We will also implement a DQN to play Atari games. Next, we will explore several interesting variants of DQN, called Double DQN, Dueling DQN, DQN with prioritized experience replay, and DRQN.

Chapter 10, Policy Gradient Method, covers policy gradient methods. We will understand how the policy gradient method works along with the detailed derivation. Next, we will learn several variance reduction methods such as policy gradient with reward-to-go and policy gradient with baseline. We will also understand how to train an agent for the Cart Pole balancing task using policy gradient.

Chapter 11, Actor-Critic Methods – A2C and A3C, deals with several interesting actor-critic methods such as advantage actor-critic and asynchronous advantage actor-critic. We will learn how these actor-critic methods work in detail, and then we will implement them for a mountain car climbing task using OpenAI Gym.

Chapter 12, Learning DDPG, TD3, and SAC, covers state-of-the-art deep RL algorithms such as deep deterministic policy gradient, twin delayed DDPG, and soft actor, along with step by step derivation. We will also learn how to implement the DDPG algorithm for performing the inverted pendulum swing-up task using Gym.

Chapter 13, TRPO, PPO, and ACKTR Methods, deals with several popular policy gradient methods such as TRPO and PPO. We will dive into the math behind TRPO and PPO step by step and understand how TRPO and PPO helps an agent find the optimal policy. Next, we will learn to implement PPO for performing the inverted pendulum swing-up task. At the end, we will learn about the actor-critic method called actor-critic using Kronecker-Factored trust region in detail.

Chapter 14, Distributional Reinforcement Learning, covers distributional RL algorithms. We will begin the chapter by understanding what distributional RL is. Then we will explore several interesting distributional RL algorithms such as categorical DQN, quantile regression DQN, and distributed distributional DDPG.

Chapter 15, Imitation Learning and Inverse RL, explains imitation and inverse RL algorithms. First, we will understand how supervised imitation learning, DAgger, and deep Q learning from demonstrations work in detail. Next, we will learn about maximum entropy inverse RL. At the end of the chapter, we will learn about generative adversarial imitation learning.

Chapter 16, Deep Reinforcement Learning with Stable Baselines, helps us to understand how to implement deep RL algorithms using a library called Stable Baselines. We will learn what Stable Baselines is and how to use it in detail by implementing several interesting Deep RL algorithms such as DQN, A2C, DDPG TRPO, and PPO.

Chapter 17, Reinforcement Learning Frontiers, covers several interesting avenues in RL, such as meta RL, hierarchical RL, and imagination augmented agents in detail.