Introduction

This chapter is the first of a series of three chapters that investigate the use of different feature sets (or spaces) in our unsupervised learning algorithms, and we will start with a discussion around dimensionality reduction, specifically, PCA. We will then extend upon our understanding of the benefits of the different feature spaces through an exploration of two independently powerful machine learning architectures in neural network-based auto-encoders. Neural networks certainly have a well-deserved reputation for being powerful models in supervised learning problems, and, through the use of an autoencoder stage, have been shown to be sufficiently flexible for their application to unsupervised learning problems. Finally, we will build on our neural network implementation and dimensionality reduction as we cover t-distributed nearest neighbors in the final chapter of this micro-series.

What Is Dimensionality Reduction?

Dimensionality reduction is an important tool in any data...