Hands-On Unsupervised Learning with Python

Implement machine learning and deep learning models using Scikit-Learn, TensorFlow, and more

Giuseppe Bonaccorso

• Éditeur: Packt Publishing
• Année: 2019
• ISBN:
  9781789349276
• Édition: 1
• Sujet: Computer Science
• Sous-sujet: Natural Language Processing

Discover the skill-sets required to implement various approaches to Machine Learning with Python

Key Features

• Explore unsupervised learning with clustering, autoencoders, restricted Boltzmann machines, and more
• Build your own neural network models using modern Python libraries
• Practical examples show you how to implement different machine learning and deep learning techniques

Book Description

Unsupervised learning is about making use of raw, untagged data and applying learning algorithms to it to help a machine predict its outcome. With this book, you will explore the concept of unsupervised learning to cluster large sets of data and analyze them repeatedly until the desired outcome is found using Python.

This book starts with the key differences between supervised, unsupervised, and semi-supervised learning. You will be introduced to the best-used libraries and frameworks from the Python ecosystem and address unsupervised learning in both the machine learning and deep learning domains. You will explore various algorithms, techniques that are used to implement unsupervised learning in real-world use cases. You will learn a variety of unsupervised learning approaches, including randomized optimization, clustering, feature selection and transformation, and information theory. You will get hands-on experience with how neural networks can be employed in unsupervised scenarios. You will also explore the steps involved in building and training a GAN in order to process images.

By the end of this book, you will have learned the art of unsupervised learning for different real-world challenges.

What you will learn

• Use cluster algorithms to identify and optimize natural groups of data
• Explore advanced non-linear and hierarchical clustering in action
• Soft label assignments for fuzzy c-means and Gaussian mixture models
• Detect anomalies through density estimation
• Perform principal component analysis using neural network models
• Create unsupervised models using GANs

Who this book is for

This book is intended for statisticians, data scientists, machine learning developers, and deep learning practitioners who want to build smart applications by implementing key building block unsupervised learning, and master all the new techniques and algorithms offered in machine learning and deep learning using real-world examples. Some prior knowledge of machine learning concepts and statistics is desirable.

Clustering Fundamentals

In this chapter, we are going to introduce the fundamental concept of cluster analysis, focusing the attention on our main principles that are shared by many algorithms and the most important techniques that can be employed to evaluate the performance of a method.

In particular, we are going to discuss:

• An introduction to clustering and distance functions
• K-means and K-means++
• Evaluation metrics
• K-Nearest Neighbors (KNN)
• Vector Quantization (VQ)

Technical requirements

The code presented in this chapter requires:

• Python 3.5+ (Anaconda distribution: https://www.anaconda.com/distribution/ is highly recommended)
• Libraries:
  • SciPy 0.19+
  • NumPy 1.10+
  • scikit-learn 0.20+
  • pandas 0.22+
  • Matplotlib 2.0+
  • seaborn 0.9+

The dataset can be obtained through UCI. The CSV file can be downloaded from https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/wdbc.data and doesn't need any preprocessing except for the addition of the column names that will occur during the loading stage.

The examples are available on the GitHub repository:

https://github.com/PacktPublishing/HandsOn-Unsupervised-Learning-with-Python/Chapter02.

Introduction to clustering

As we explained in Chapter 1, Getting Started with Unsupervised Learning, the main goal of a cluster analysis is to group the elements of a dataset according to a similarity measure or a proximity criterion. In the first part of this chapter, we are going to focus on the former approach, while in the second part and in the next chapter, we will analyze more generic methods that exploit other geometric features of the dataset.

Let's take a data generating process p_data(x) and draw N samples from it:

It's possible to assume that the probability space of p_data(x) is partitionable into (potentially infinite) configurations containing K (for K=1,2, ...) regions so that p_data(x; k) represents the probability of a sample belonging to a cluster k. In this way, we are stating that every possible clustering structure is already existing when p_data(x) is determined. Further assumptions can be made on the clustering probability distribution that better approximate p_data(x) (as we're going to see in Chapter 5, Soft Clustering and Gaussian Mixture Models). However, as we are trying to split the probability space (and the corresponding samples) into cohesive groups, we can assume two possible strategies:

• Hard clustering: In this case, each sample x_p ∈ X is assigned to a cluster K_i and K_i ∩ K_j = ∅ for i ≠ j. The majority of algorithms we are going to discuss belong to this category. In this case, the problem can be expressed as a parameterized function that assigns a cluster to each input sample:

• Soft clustering: It is often subdivided into probabilistic and fuzzy clustering and such an approach determines the probability p(x) of every sample x_p ∈ X belonging to predetermined clusters. Hence, if there are K clusters, we have a probability vector p(x) = [p₁(x), p₂(x), ..., p_k(x)], where p_i(x) represents the probability of being assigned to the cluster i. In this case, the clusters are not disjointed and, generally, a sample will belong to all clusters with a membership degree that is equivalent to a probability (this concept is peculiar to fuzzy clustering).

For our purposes, in this chapter we simply assume that the dataset X is drawn from a data generating process whose space, given a metric function, is splittable into compact regions separated from each other. In fact, our main objective is to find K clusters that satisfy the double property of maximum cohesion and maximum separation. This concept will be clearer when discussing the K-means algorithm. However, it's possible to imagine clusters as blobs whose density is much higher than the one observable in the space separating two or more of them, as shown in the following diagram:

Bidimensional clustering structure obeying the rule of maximum cohesion and maximum separation. N_k represents the number of samples belonging to the cluster k while N_out(r) is the number of samples that are outside the balls centered at each cluster center with a maximum radius r

In the preceding diagram, we are assuming that the majority of samples will be captured by one of the balls, considering the maximum distance of a sample from the center. However, as we don't want to impose any restriction on the growth of a ball (that is, it can contain any number of samples), it's preferable not to consider the radius and to evaluate the separating region by sampling small subregions (of the whole space) and collecting their densities.

In a perfect scenario, the clusters span some subregions whose density is D, while the separating region is characterized by a density d << D. The discussion about geometric properties can become extremely complex and, in many cases, it's extremely theoretical. Henceforth, we consider only the distance between the closest points belonging to different clusters. If this value is much smaller than the maximum distance between a sample and its cluster center for all clusters, we can be sure that the separation is effective and it's easy to distinguish between clusters and separating regions. Instead, when...