Machine Learning For Dummies

Chapter 12

Starting with Simple Learners

IN THIS CHAPTER

Trying a perceptron to separate classes by a line

Partitioning recursively training data by decision trees

Discovering the rules behind playing tennis and surviving the Titanic

Leveraging Bayesian probability to analyze textual data

Beginning with this chapter, the examples start illustrating the basics of how to learn from data. The plan is to touch some of the simplest learning strategies first — providing some formulas (just those that are essential), intuitions about their functioning, and examples in R and Python for experimenting with some of their most typical characteristics. The chapter begins by reviewing the use of the perceptron to separate classes.

At the root of all principal machine learning techniques presented in the book, there is always an algorithm based on somewhat interrelated linear combinations, variations of the sample splitting of decision trees, or some kind of Bayesian probabilistic reasoning. This chapter uses classification trees to demonstrate the technique. The only exception is the K-Nearest Neighbors (KNN) algorithm, which, based on analogical reasoning, is treated apart in a special chapter devoted to detection of similarity in data (Chapter 14).

Getting a grasp on these basic techniques means being able to deal with more complex learning techniques later and being able to understand (and use) them better. It may appear incredible now, but you can create some of the most effective algorithms ...