In this chapter we will review the fundamental concepts in machine learning. We will discuss applications of machine learning algorithms, the supervised-unsupervised learning spectrum, uses of training and testing data, and model evaluation. Finally, we will introduce scikit-learn, and install the tools required in subsequent chapters.
Our imagination has long been captivated by visions of machines that can learn and imitate human intelligence. While visions of general artificial intelligence such as Arthur C. Clarke's HAL and Isaac Asimov's Sonny have yet to be realized, software programs that can acquire new knowledge and skills through experience are becoming increasingly common. We use such machine learning programs to discover new music that we enjoy, and to quickly find the exact shoes we want to purchase online. Machine learning programs allow us to dictate commands to our smartphones and allow our thermostats to set their own temperatures. Machine learning programs can decipher sloppily-written mailing addresses better than humans, and guard credit cards from fraud more vigilantly. From investigating new medicines to estimating the page views for versions of a headline, machine learning software is becoming central to many industries. Machine learning has even encroached on activities that have long been considered uniquely human, such as writing the sports column recapping the Duke basketball team's loss to UNC.
Machine learning is the design and study of software artifacts that use past experience to make future decisions; it is the study of programs that learn from data. The fundamental goal of machine learning is to generalize, or to induce an unknown rule from examples of the rule's application. The canonical example of machine learning is spam filtering. By observing thousands of emails that have been previously labeled as either spam or ham, spam filters learn to classify new messages.
Arthur Samuel, a computer scientist who pioneered the study of artificial intelligence, said that machine learning is "the study that gives computers the ability to learn without being explicitly programmed." Throughout the 1950s and 1960s, Samuel developed programs that played checkers. While the rules of checkers are simple, complex strategies are required to defeat skilled opponents. Samuel never explicitly programmed these strategies, but through the experience of playing thousands of games, the program learned complex behaviors that allowed it to beat many human opponents.
A popular quote from computer scientist Tom Mitchell defines machine learning more formally: "A program can be said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E." For example, assume that you have a collection of pictures. Each picture depicts either a dog or cat. A task could be sorting the pictures into separate collections of dog and cat photos. A program could learn to perform this task by observing pictures that have already been sorted, and it could evaluate its performance by calculating the percentage of correctly classified pictures.
We will use Mitchell's definition of machine learning to organize this chapter. First, we will discuss types of experience, including supervised learning and unsupervised learning. Next, we will discuss common tasks that can be performed by machine learning systems. Finally, we will discuss performance measures that can be used to assess machine learning systems.
Machine learning systems are often described as learning from experience either with or without supervision from humans. In supervised learning problems, a program predicts an output for an input by learning from pairs of labeled inputs and outputs; that is, the program learns from examples of the right answers. In unsupervised learning, a program does not learn from labeled data. Instead, it attempts to discover patterns in the data. For example, assume that you have collected data describing the heights and weights of people. An example of an unsupervised learning problem is dividing the data points into groups. A program might produce groups that correspond to men and women, or children and adults.
Now assume that the data is also labeled with the person's sex. An example of a supervised learning problem is inducing a rule to predict whether a person is male or female based on his or her height and weight. We will discuss algorithms and examples of supervised and unsupervised learning in the following chapters.
Supervised learning and unsupervised learning can be thought of as occupying opposite ends of a spectrum. Some types of problems, called semi-supervised learning problems, make use of both supervised and unsupervised data; these problems are located on the spectrum between supervised and unsupervised learning. An example of semi-supervised machine learning is reinforcement learning, in which a program receives feedback for its decisions, but the feedback may not be associated with a single decision. For example, a reinforcement learning program that learns to play a side-scrolling video game such as Super Mario Bros. may receive a reward when it completes a level or exceeds a certain score, and a punishment when it loses a life. However, this supervised feedback is not associated with specific decisions to run, avoid Goombas, or pick up fire flowers. While this book will discuss semi-supervised learning, we will focus primarily on supervised and unsupervised learning, as these categories include most the common machine learning problems. In the next sections, we will review supervised and unsupervised learning in more detail.
A supervised learning program learns from labeled examples of the outputs that should be produced for an input. There are many names for the output of a machine learning program. Several disciplines converge in machine learning, and many of those disciplines use their own terminology. In this book, we will refer to the output as the response variable. Other names for response variables include dependent variables, regressands, criterion variables, measured variables, responding variables, explained variables, outcome variables, experimental variables, labels, and output variables. Similarly, the input variables have several names. In this book, we will refer to the input variables as features, and the phenomena they measure as explanatory variables. Other names for explanatory variables include predictors, regressors, controlled variables, manipulated variables, and exposure variables. Response variables and explanatory variables may take real or discrete values.
The collection of examples that comprise supervised experience is called a training set. A collection of examples that is used to assess the performance of a program is called a test set. The response variable can be thought of as the answer to the question posed by the explanatory variables. Supervised learning problems learn from a collection of answers to different questions; that is, supervised learning programs are provided with the correct answers and must learn to respond correctly to unseen, but similar, questions.