Book Image

The Unsupervised Learning Workshop

By : Aaron Jones, Christopher Kruger, Benjamin Johnston
Book Image

The Unsupervised Learning Workshop

By: Aaron Jones, Christopher Kruger, Benjamin Johnston

Overview of this book

Do you find it difficult to understand how popular companies like WhatsApp and Amazon find valuable insights from large amounts of unorganized data? The Unsupervised Learning Workshop will give you the confidence to deal with cluttered and unlabeled datasets, using unsupervised algorithms in an easy and interactive manner. The book starts by introducing the most popular clustering algorithms of unsupervised learning. You'll find out how hierarchical clustering differs from k-means, along with understanding how to apply DBSCAN to highly complex and noisy data. Moving ahead, you'll use autoencoders for efficient data encoding. As you progress, you’ll use t-SNE models to extract high-dimensional information into a lower dimension for better visualization, in addition to working with topic modeling for implementing natural language processing (NLP). In later chapters, you’ll find key relationships between customers and businesses using Market Basket Analysis, before going on to use Hotspot Analysis for estimating the population density of an area. By the end of this book, you’ll be equipped with the skills you need to apply unsupervised algorithms on cluttered datasets to find useful patterns and insights.
Table of Contents (11 chapters)
Preface

t-Distributed SNE

t-SNE aims to address the crowding problem using a modified version of the KL divergence cost function and by substituting the Gaussian distribution with the Student's t-distribution in the low-dimensional space. The Student's t-distribution is a probability distribution much like Gaussian and is used when we have a small sample size and unknown population standard deviation. It is often used in the Student's t-test.

The modified KL cost function considers the pairwise distances in the low-dimensional space equally, while the Student's distribution employs a heavy tail in the low-dimensional space to avoid the crowding problem. In the higher-dimensional probability calculation, the Gaussian distribution is still used to ensure that a moderate distance in the higher dimensions is still represented as such in the lower dimensions. This combination of different distributions in the respective spaces allows for the faithful representation of datapoints...