Tidy Modeling with R: A Framework for Modeling in the Tidyverse

384 Pages·2022·21.39 MB·English

by  Max Kuhn, Julia Silge

Tidy Modeling
with R
A Framework for Modeling in the Tidyverse
K
u
h
n
&
S
ilg
e
Max Kuhn & Julia Silge
Tidy Modeling with R
Get going with tidymodels, a collection of R packages for
modeling and machine learning. Whether you’re just starting “The tidymodels
out or have years of experience with modeling, this practical framework combines
introduction shows data analysts, business analysts, and data human-centric design
scientists how the tidymodels framework offers a consistent, with statistical best
flexible approach for your work.
practices, and I can’t
RStudio engineers Max Kuhn and Julia Silge demonstrate think of a better way to
ways to create models by focusing on an R dialect called the learn it than from Max
tidyverse. Software that adopts tidyverse principles shares and Julia.”
both a high-level design philosophy and low-level grammar
— Hadley Wickham
and data structures, so learning one piece of the ecosystem Chief Scientist, RStudio
makes it easier to learn the next. You’ll understand why the
tidymodels framework has been built to be used by a broad “This book offers a unified
range of people. and systematic approach
to building, analyzing,
With this book, you will:
and evaluating statistical
• Learn the steps necessary to build a model from beginning
models in R.”
to end
— Balasubramanian Narasimhan
• Understand how to use different modeling and feature Senior Research Scientist,
engineering approaches fluently Stanford University
• Examine the options for avoiding common pitfalls of
modeling, such as overfitting Max Kuhn is a software engineer at
RStudio, where he works to improve
• Learn practical methods to prepare your data for modeling
R’s modeling capabilities. He’s applied
• Tune models for optimal performance models in the pharmaceutical and
diagnostic industries for more than
• Use good statistical practices to compare, evaluate, and 18 years.
choose among models K
Julia Silge is a software engineer at u
h
RStudio working on open source n
modeling tools. With a PhD in &
astrophysics, she’s worked as a data S
scientist in tech and the nonprofit sector. ilg
e
DATA SCIENCE Twitter: @oreillymedia
linkedin.com/company/oreilly-media
US $59.99 CAN $74.99 youtube.com/oreillymedia
ISBN: 978-1-492-09648-1
Tidy Modeling with R
A Framework for Modeling in the Tidyverse
Max Kuhn and Julia Silge
BBeeiijjiinngg BBoossttoonn FFaarrnnhhaamm SSeebbaassttooppooll TTookkyyoo
Tidy Modeling with R
by Max Kuhn and Julia Silge
Copyright © 2022 Max Kuhn and Julia Silge. All rights reserved.
Printed in the United States of America.
Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472.
O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are
also available for most titles (http://oreilly.com). For more information, contact our corporate/institutional
sales department: 800-998-9938 or [email protected].
Acquisitions Editor: Michelle Smith Indexer: Potomac Indexing, LLC
Development Editor: Rita Fernando Interior Designer: David Futato
Production Editor: Beth Kelly Cover Designer: Karen Montgomery
Copyeditor: Piper Editorial Consulting, LLC Illustrator: Kate Dullea
Proofreader: Tom Sullivan
July 2022: First Edition
Revision History for the First Edition
2022-07-12: First Release
See http://oreilly.com/catalog/errata.csp?isbn=9781492096481 for release details.
The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. Tidy Modeling with R, the cover image,
and related trade dress are trademarks of O’Reilly Media, Inc.
The views expressed in this work are those of the authors and do not represent the publisher’s views.
While the publisher and the authors have used good faith efforts to ensure that the information and
instructions contained in this work are accurate, the publisher and the authors disclaim all responsibility
for errors or omissions, including without limitation responsibility for damages resulting from the use
of or reliance on this work. Use of the information and instructions contained in this work is at your
own risk. If any code samples or other technology this work contains or describes is subject to open
source licenses or the intellectual property rights of others, it is your responsibility to ensure that your use
thereof complies with such licenses and/or rights.
978-1-492-09648-1
[LSI]
To Amy: When you read this, know that I love you more today than every day before.
—M.K.
To Robert: Happy 20 years of choosing each other.
—J.S.

Table of Contents
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
Part I. Introduction
1. Software for Modeling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Fundamentals for Modeling Software 4
Types of Models 5
Descriptive Models 6
Inferential Models 8
Predictive Models 9
Connections Between Types of Models 10
Some Terminology 11
How Does Modeling Fit into the Data Analysis Process? 12
Chapter Summary 15
2. A Tidyverse Primer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Tidyverse Principles 17
Design for Humans 18
Reuse Existing Data Structures 19
Design for the Pipe and Functional Programming 20
Examples of Tidyverse Syntax 23
Chapter Summary 25
3. A Review of R Modeling Fundamentals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
An Example 27
What Does the R Formula Do? 33
Why Tidiness Is Important for Modeling 34
vi
Combining Base R Models and the Tidyverse 38
The tidymodels Metapackage 39
Chapter Summary 41
Part II. Modeling Basics
4. The Ames Housing Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Exploring Features of Homes in Ames 47
Chapter Summary 52
5. Spending Our Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Common Methods for Splitting Data 56
What About a Validation Set? 59
Multilevel Data 59
Other Considerations for a Data Budget 60
Chapter Summary 61
6. Fitting Models with parsnip. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Create a Model 63
Use the Model Results 69
Make Predictions 71
parsnip-Extension Packages 73
Creating Model Specifications 73
Chapter Summary 74
7. A Model Workflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Where Does the Model Begin and End? 75
Workflow Basics 78
Adding Raw Variables to the workflow() 80
How Does a workflow() Use the Formula? 81
Tree-Based Models 82
Special Formulas and Inline Functions 82
Creating Multiple Workflows at Once 84
Evaluating the Test Set 86
Chapter Summary 87
8. Feature Engineering with Recipes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
A Simple recipe() for the Ames Housing Data 90
Using Recipes 92
How Data Are Used by the recipe() 94
Examples of Steps 94
vii | Table of Contents
Encoding Qualitative Data in a Numeric Format 95
Interaction Terms 97
Spline Functions 99
Feature Extraction 101
Row Sampling Steps 102
General Transformations 103
Natural Language Processing 103
Skipping Steps for New Data 103
Tidy a recipe() 104
Column Roles 106
Chapter Summary 107
9. Judging Model Effectiveness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Performance Metrics and Inference 110
Regression Metrics 112
Binary Classification Metrics 114
Multiclass Classification Metrics 117
Chapter Summary 122
Part III. Tools for Creating Effective Models
10. Resampling for Evaluating Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
The Resubstitution Approach 125
Resampling Methods 128
Cross-Validation 129
Repeated Cross-Validation 132
Leave-One-Out Cross-Validation 133
Monte Carlo Cross-Validation 133
Validation Sets 134
Bootstrapping 135
Rolling Forecasting Origin Resampling 136
Estimating Performance 138
Parallel Processing 143
Saving the Resampled Objects 144
Chapter Summary 147
11. Comparing Models with Resampling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Creating Multiple Models with Workflow Sets 149
Comparing Resampled Performance Statistics 152
Simple Hypothesis Testing Methods 155
Bayesian Methods 157
Table of Contents | viii
A Random Intercept Model 158
The Effect of the Amount of Resampling 163
Chapter Summary 164
12. Model Tuning and the Dangers of Overfitting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Model Parameters 165
Tuning Parameters for Different Types of Models 166
What Do We Optimize? 168
The Consequences of Poor Parameter Estimates 173
Two General Strategies for Optimization 176
Tuning Parameters in tidymodels 177
Chapter Summary 183
13. Grid Search. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Regular and Nonregular Grids 185
Regular Grids 186
Nonregular Grids 188
Evaluating the Grid 191
Finalizing the Model 196
Tools for Creating Tuning Specifications 198
Tools for Efficient Grid Search 199
Submodel Optimization 199
Parallel Processing 200
Benchmarking Boosted Trees 204
Access to Global Variables 206
Racing Methods 208
Chapter Summary 210
14. Iterative Search. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
A Support Vector Machine Model 213
Bayesian Optimization 216
A Gaussian Process Model 216
Acquisition Functions 218
The tune_bayes() Function 222
Simulated Annealing 227
Simulated Annealing Search Process 227
The tune_sim_anneal() Function 230
Chapter Summary 234
15. Screening Many Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Modeling Concrete Mixture Strength 235
Creating the Workflow Set 238
ix | Table of Contents