Chapter 1. What is machine learning?

This chapter covers

  • Machine-learning basics
  • Advantages of machine learning over traditional approaches
  • Overview of the basic machine-learning workflow
  • Overview of advanced methods for improving model performance

In 1959, an IBM computer scientist named Arthur Samuel wrote a computer program to play checkers. Each board position was assigned a score based on its likelihood of leading to a win. At first, scores were based on a formula using factors such as the number of pieces on each side and the number of kings. It worked, but Samuel had an idea about how to improve its performance. He had the program play thousands of games against itself and used the results to refine the positional scoring. By the mid-1970s, the program had achieved the proficiency of a respectable amateur player.[1]

1Jonathan Schaeffer, One Jump Ahead: Computer Perfection at Checkers (New York: Springer, 2009).

Samuel had written a computer program that was able to improve its own performance through experience. It learned—and machine learning (ML) was born.

1.1. Understanding how machines learn

1.2. Using data to make decisions

1.3. Following the ML workflow: from data to deployment

1.4. Boosting model performance with advanced techniques

1.5. Summary

1.6. Terms from this chapter



instance or example A single object, observation, transaction, or record.
target or label The numerical or categorical (label) attribute of interest. This is the variable to be predicted for each new instance.
features The input attributes that are used to predict the target. These also may be numerical or categorical.
model A mathematical object describing the relationship between the features and the target.
training data The set of instances with a known target to be used to fit an ML model.
recall Using a model to predict a target or label.
supervised machine learning Machine learning in which, given examples for which the output value is known, the training process infers a function that relates input values to the output.
unsupervised machine learning Machine-learning techniques that don’t rely on labeled examples, but rather try to find hidden structure in unlabeled data.
ML workflow The stages in the ML process: data preparation, model building, evaluation, optimization, and prediction.
online machine learning A form of machine learning in which predictions are made, and the model is updated, for each new example.