CHAPTER 1
Introduction to Machine Learning

Welcome to Python Machine Learning! The fact that you are reading this book is a clear indication of your interest in this very interesting and exciting topic.

This book covers machine learning, one of the hottest programming topics in more recent years. Machine learning (ML) is a collection of algorithms and techniques used to design systems that learn from data. These systems are then able to perform predictions or deduce patterns from the supplied data.

With computing power increasing exponentially and prices decreasing simultaneously, there is no better time for machine learning. Machine learning tasks that usually require huge processing power are now possible on desktop machines. Nevertheless, machine learning is not for the faint of heart-it requires a good foundation in mathematics, statistics, as well as programming knowledge. The majority of the books in the market on machine learning go into too much detail, which often leaves beginning readers gasping for air. Most of the discussion on machine learning revolves heavily around statistical theories and algorithms, so unless you are a mathematician or a PhD candidate, you will likely find them difficult to digest. For most people, developers in particular, what they want is to have a foundational understanding of how machine learning works, and most importantly, how to apply machine learning in their applications. It is with this motive in mind that I set out to write this book.

This book will take a gentle approach to machine learning. I will attempt to do the following:

• Cover the libraries in Python that lay the foundation for machine learning, namely NumPy, Pandas, and matplotlib.
• Discuss machine learning using Python and the Scikit-learn libraries. Where possible, I will manually implement the relevant machine learning algorithm using Python. This will allow you to understand how the various machine learning algorithms work behind the scenes. Once this is done, I will show how to use the Scikit-learn libraries, which make it really easy to integrate machine learning into your own apps.
• Cover the common machine learning algorithms-regressions, clustering, and classifications.

TIP

It is not the intention of this book to go into a deep discussion of machine learning algorithms. Although there are chapters that discuss some of the mathematical concepts behind the algorithms, it is my intention to make the subject easy to understand and hopefully motivate you to learn further.

Machine learning is indeed a very complex topic. But instead of discussing the complex mathematical theories behind it, I will cover it using easy-to-understand examples and walk you through numerous code samples. This code-intensive book encourages readers to try out the numerous examples in the various chapters, which are designed to be independent, compact, and easy to follow and understand.

What Is Machine Learning?

If you have ever written a program, you will be familiar with the diagram shown in Figure 1.1. You write a program, feed some data into it, and get your output. For example, you might write a program to perform some accounting tasks for your business. In this case, the data collected would include your sales records, your inventory lists, and so on. The program would then take in the data and calculate your profits or loss based on your sales records. You may also perhaps churn out some nice and fanciful charts showing your sales performance. In this case, the output is the profit/loss statement, as well as other charts.

Figure 1.1: In traditional programming, the data and the program produce the output

For many years, traditional desktop and web programming have dominated the landscape, and many algorithms and methodologies have evolved to make programs run more efficiently. In more recent years, however, machine learning has taken over the programming world. Machine learning has transformed the paradigm in Figure 1.1 to a new paradigm, which is shown in Figure 1.2. Instead of feeding the data to the program, you now use the data and the output that you have collected to derive your program (also known as the model). Using the same accounting example, with the machine learning paradigm, you would take the detailed sales records (which are collectively both the data and output) and use them to derive a set of rules to make predictions. You may use this model to predict the most popular items that will sell next year, or which items will be less popular going forward.

Figure 1.2: In machine learning, the data and the output produce the program

TIP

Machine learning is about finding patterns in data.

What Problems Will Machine Learning Be Solving in This Book?

So, what exactly is machine learning? Machine learning (ML) is a collection of algorithms and techniques used to design systems that learn from data. ML algorithms have a strong mathematical and statistical basis, but they do not take into account domain knowledge. ML consists of the following disciplines:

• Scientific computing
• Mathematics
• Statistics

A good application of machine learning is trying to determine if a particular credit card transaction is fraudulent. Given past transaction records, the data scientist's job is to clean up and transform the data based on domain knowledge so that the right ML algorithm can be applied in order to solve the problem (in this case determine if a transaction is fraudulent). A data scientist needs to know about which method of machine learning will best help in completing this task and how to apply it. The data scientist does not necessarily need to know how that method works, although knowing this will always help in building a more accurate learning model.

In this book, there are three main types of problems that we want to solve using machine learning. These problem types are as follows:

• Classification: Is this A or B?
• Regression: How much or how many?
• Clustering: How is this organized?

Classification

In machine learning, classification is identifying to which set of categories a new observation belongs based on the set of training data containing in the observed categories. Here are some examples of classification problems:

• Predicting the winner for the U.S. 2020 Presidential Election
• Predicting if a tumor is cancerous
• Classifying the different types of flowers

A classification problem with two classes is known as a two-class classification problem. Those with more than two classes are known as multi-class classification problems.

The outcome of a classification problem is a discrete value indicating the predicted class in which an observation lies. The outcome of a classification problem can also be a continuous value, indicating the likelihood of an observation belonging to a particular class. For example, candidate A is predicted to win the election with a probability of 0.65 (or 65 percent). Here, 0.65 is the continuous value indicating the confidence of the prediction, and it can be converted to a class value ("win" in this case) by selecting the prediction with the highest probability.

Chapter 7 through Chapter 9 will discuss classifications in more detail.

Regression

Regression helps in forecasting the future by estimating the relationship between variables. Unlike classification (which predicts the class to which an observation belongs), regression returns a continuous output variable. Here are some examples of regression problems:

• Predicting the sales number for a particular item for the next quarter
• Predicting the temperatures for next week
• Predicting the lifespan of a particular model of tire

Chapter 5 and Chapter 6 will discuss regressions in more detail.

Clustering

Clustering helps in grouping similar data points into intuitive groups. Given a set of data, clustering helps you discover how they are organized by grouping them into natural clumps.

Examples of clustering problems are as follows:

• Which viewers like the same genre of movies
• Which models of hard drives fail in the same way

Clustering is very useful when you want to discover a specific pattern in the data. Chapter 10 will discuss clustering in more detail.

Types of Machine Learning Algorithms

Machine learning algorithms fall into two broad categories:

• Supervised learning algorithms are trained with labeled data. In other words, data composed of examples of the desired answers. For instance, a model that identifies fraudulent credit card use would be trained from a dataset with labeled data points of known fraudulent and valid charges. Most machine learning is supervised.
• Unsupervised learning algorithms are used on data with no labels, and the goal is to find relationships in the data. For instance, you might want to find groupings of customer demographics with similar buying habits.

Supervised Learning

In supervised learning, a labeled dataset is used. A labeled dataset means that a group of data...