What this book covers
Chapter 1, Giving Computers the Ability to Learn from Data, introduces the main subareas of machine learning used to tackle various problem tasks. In addition, it discusses the essential steps for creating a typical machine learning model-building pipeline that will guide us through the following chapters.
Chapter 2, Training Simple Machine Learning Algorithms for Classification, goes back to the origin of machine learning and introduces binary perceptron classifiers and adaptive linear neurons. This chapter is a gentle introduction to the fundamentals of pattern classification and focuses on the interplay of optimization algorithms and machine learning.
Chapter 3, A Tour of Machine Learning Classifiers Using scikit-learn, describes the essential machine learning algorithms for classification and provides practical examples using one of the most popular and comprehensive open source machine learning libraries, scikit-learn.
Chapter 4, Building Good Training Datasets – Data Preprocessing, discusses how to deal with the most common problems in unprocessed datasets, such as missing data. It also discusses several approaches to identify the most informative features in datasets and how to prepare variables of different types as proper inputs for machine learning algorithms.
Chapter 5, Compressing Data via Dimensionality Reduction, describes the essential techniques to reduce the number of features in a dataset to smaller sets, while retaining most of their useful and discriminatory information. It also discusses the standard approach to dimensionality reduction via principal component analysis and compares it to supervised and nonlinear transformation techniques.
Chapter 6, Learning Best Practices for Model Evaluation and Hyperparameter Tuning, discusses the dos and don'ts for estimating the performance of predictive models. Moreover, it discusses different metrics for measuring the performance of our models and techniques for fine-tuning machine learning algorithms.
Chapter 7, Combining Different Models for Ensemble Learning, introduces the different concepts of combining multiple learning algorithms effectively. It explores how to build ensembles of experts to overcome the weaknesses of individual learners, resulting in more accurate and reliable predictions.
Chapter 8, Applying Machine Learning to Sentiment Analysis, discusses the essential steps for transforming textual data into meaningful representations for machine learning algorithms to predict the opinions of people based on their writing.
Chapter 9, Embedding a Machine Learning Model into a Web Application, continues with the predictive model from the previous chapter and walks through the essential steps of developing web applications with embedded machine learning models.
Chapter 10, Predicting Continuous Target Variables with Regression Analysis, discusses the essential techniques for modeling linear relationships between target and response variables to make predictions on a continuous scale. After introducing different linear models, it also talks about polynomial regression and tree-based approaches.
Chapter 11, Working with Unlabeled Data – Clustering Analysis, shifts the focus to a different subarea of machine learning, unsupervised learning. It covers algorithms from three fundamental families of clustering algorithms that find groups of objects that share a certain degree of similarity.
Chapter 12, Implementing a Multilayer Artificial Neural Network from Scratch, extends the concept of gradient-based optimization, which we first introduced in Chapter 2, Training Simple Machine Learning Algorithms for Classification. In this chapter, we will build powerful, multilayer neural networks (NNs) based on the popular backpropagation algorithm in Python.
Chapter 13, Parallelizing Neural Network Training with TensorFlow, builds upon the knowledge from the previous chapter to provide a practical guide for training NNs more efficiently. The focus of this chapter is on TensorFlow 2.0, an open source Python library that allows us to utilize multiple cores of modern graphics processing units (GPUs) and construct deep NNs from common building blocks via the user-friendly Keras API.
Chapter 14, Going Deeper – The Mechanics of TensorFlow, picks up where the previous chapter left off and introduces the more advanced concepts and functionality of TensorFlow 2.0. TensorFlow is an extraordinarily vast and sophisticated library, and this chapter walks through concepts such as compiling code into a static graph for faster execution and defining trainable model parameters. In addition, this chapter provides additional hands-on experience of training deep neural networks using TensorFlow's Keras API, as well as TensorFlow's pre-made Estimators.
Chapter 15, Classifying Images with Deep Convolutional Neural Networks, introduces convolutional neural networks (CNNs). A CNN represents a particular type of deep NN architecture that is particularly well suited for image datasets. Due to their superior performance compared to traditional approaches, CNNs are now widely used in computer vision to achieve state-of-the-art results for various image recognition tasks. Throughout this chapter, you will learn how convolutional layers can be used as powerful feature extractors for image classification.
Chapter 16, Modeling Sequential Data Using Recurrent Neural Networks, introduces another popular NN architecture for deep learning that is especially well suited to working with text and other types of sequential data and time series data. As a warm-up exercise, this chapter introduces recurrent NNs for predicting the sentiment of movie reviews. Then, the chapter covers teaching recurrent networks to digest information from books in order to generate entirely new text.
Chapter 17, Generative Adversarial Networks for Synthesizing New Data, introduces a popular adversarial training regime for NNs that can be used to generate new, realistic-looking images. The chapter starts with a brief introduction to autoencoders, a particular type of NN architecture that can be used for data compression. The chapter then shows how to combine the decoder part of an autoencoder with a second NN that can distinguish between real and synthesized images. By letting two NNs compete with each other in an adversarial training approach, you will implement a generative adversarial network that generates new handwritten digits. Lastly, after introducing the basic concepts of generative adversarial networks, the chapter introduces improvements that can stabilize the adversarial training, such as using the Wasserstein distance metric.
Chapter 18, Reinforcement Learning for Decision Making in Complex Environments, covers a subcategory of machine learning that is commonly used for training robots and other autonomous systems. This chapter starts by introducing the basics of reinforcement learning (RL) to make you familiar with agent/environment interactions, the reward process of RL systems, and the concept of learning from experience. The chapter covers the two main categories of RL, model-based and model-free RL. After learning about basic algorithmic approaches, such as Monte Carlo- and temporal distance-based learning, you will implement and train an agent that can navigate a grid world environment using the Q-learning algorithm.
Finally, this chapter introduces the deep Q-learning algorithm, which is a variant of Q-learning that uses deep NNs.