Chapter 5. Learning and Prediction
In this chapter, we’ll learn what our data means and how it drives our decision processes. Learning about our data gives us knowledge, and knowledge enables us to make reasonable guesses about what to expect in the future. This is the reason for the existence of data science: learning enough about the data so we can make predictions on newly arriving data. This can be as simple as categorizing data into groups or clusters. It can span a much broader set of processes that culminate (ultimately) in the path to artificial intelligence. Learning is divided into two major categories: unsupervised and supervised.
In general, we think of data as having variates X and responses Y, and our goal is to build a model using X so that we can predict what happens when we put in a new X. If we have the Y, we can “supervise” the building of the model. In many cases, we have only the variates X. The model will then have to be built in an unsupervised manner. Typical unsupervised methods include clustering, whereas supervised learning may include any of the regression methods (e.g., linear regression) or classification methods such as naive Bayes, logistic, or deep neural net classifiers. Many other methods and permutations of those methods exist, and covering them all would be impossible. Instead, here we dive into a few of the most useful ones.
A few learning algorithms are prevalent in a large variety of techniques. In particular, we often use ...