Book Image

Hands-On Ensemble Learning with Python

By : George Kyriakides, Konstantinos G. Margaritis

Book Image

Hands-On Ensemble Learning with Python

By: George Kyriakides, Konstantinos G. Margaritis

Overview of this book

Ensembling is a technique of combining two or more similar or dissimilar machine learning algorithms to create a model that delivers superior predictive power. This book will demonstrate how you can use a variety of weak algorithms to make a strong predictive model. With its hands-on approach, you'll not only get up to speed with the basic theory but also the application of different ensemble learning techniques. Using examples and real-world datasets, you'll be able to produce better machine learning models to solve supervised learning problems such as classification and regression. In addition to this, you'll go on to leverage ensemble learning techniques such as clustering to produce unsupervised machine learning models. As you progress, the chapters will cover different machine learning algorithms that are widely used in the practical world to make predictions and classifications. You'll even get to grips with the use of Python libraries such as scikit-learn and Keras for implementing different ensemble models. By the end of this book, you will be well-versed in ensemble learning, and have the skills you need to understand which ensemble method is required for which problem, and successfully implement them in real-world scenarios.

Preface

Who this book is for

What this book covers

To get the most out of this book

Free Chapter

Section 1: Introduction and Required Software Tools

Section 1: Introduction and Required Software Tools

A Machine Learning Refresher

A Machine Learning Refresher

Technical requirements

Learning from data

Supervised and unsupervised learning

Performance measures

Machine learning algorithms

Getting Started with Ensemble Learning

Getting Started with Ensemble Learning

Technical requirements

Bias, variance, and the trade-off

Ensemble learning

Difficulties in ensemble learning

Section 2: Non-Generative Methods

Section 2: Non-Generative Methods

Voting

Technical requirements

Hard and soft voting

Python implementation

Using scikit-learn

Stacking

Technical requirements

Deciding on an ensemble's composition

Python implementation

Section 3: Generative Methods

Section 3: Generative Methods

Bagging

Technical requirements

Python implementation

Using scikit-learn

Boosting

Technical requirements

Gradient boosting

Using scikit-learn

Random Forests

Technical requirements

Understanding random forest trees

Creating forests

Using scikit-learn

Section 4: Clustering

Section 4: Clustering

Clustering

Technical requirements

Consensus clustering

Using OpenEnsembles

Section 5: Real World Applications

Section 5: Real World Applications

Classifying Fraudulent Transactions

Classifying Fraudulent Transactions

Technical requirements

Getting familiar with the dataset

Exploratory analysis

Using random forests

Comparative analysis of ensembles

Predicting Bitcoin Prices

Predicting Bitcoin Prices

Technical requirements

Time series data

Evaluating Sentiment on Twitter

Evaluating Sentiment on Twitter

Technical requirements

Sentiment analysis tools

Getting Twitter data

Creating a model

Classifying tweets in real time

Recommending Movies with Keras

Recommending Movies with Keras

Technical requirements

Demystifying recommendation systems

Neural recommendation systems

Using Keras for movie recommendations

Clustering World Happiness

Clustering World Happiness

Technical requirements

Understanding the World Happiness Report

Creating the ensemble

Gaining insights

Another Book You May Enjoy

Another Book You May Enjoy

Leave a review - let other readers know what you think

Customer Reviews

5 star

0

4 star

0

3 star

0

2 star

0

1 star

0

Summary

In this chapter, we tried to model historical Bitcoin prices using all of the ensemble methods presented in earlier chapters of this book. As with most datasets, there are many decisions that affect a model's quality. Data preprocessing and feature engineering are some of the most important factors, especially when the dataset's nature does not allow direct modeling of the data. Time series datasets fall into this category, in which the construction of appropriate features and targets is required. By transforming our non-stationary time series to stationary, we improved the algorithm's ability to model the data.

To assess the quality of our models, we used the MSE of return percentages, as well as the Sharpe ratio (in which we assumed that the model was utilized as a trading strategy). When MSE is concerned, the best performing ensemble proved to be the...