FRANCESCA LAZZERI is an accomplished economist who works with machine learning, artificial intelligence, and applied econometrics. She works at Microsoft as a data scientist and machine learning scientist to develop a portfolio of machine learning services. She is a sought-after speaker and has given popular talks at AI conferences and academic seminars at Berkeley, Harvard, and MIT.

Acknowledgments vii

Introduction xv

Chapter 1 Overview of Time Series Forecasting 1

Flavors of Machine Learning for Time Series Forecasting 3

Supervised Learning for Time Series Forecasting 14

Python for Time Series Forecasting 21

Experimental Setup for Time Series Forecasting 24

Conclusion 26

Chapter 2 How to Design an End-to-End Time Series Forecasting Solution on the Cloud 29

Time Series Forecasting Template 31

Business Understanding and Performance Metrics 33

Data Ingestion 36

Data Exploration and Understanding 39

Data Pre-processing and Feature Engineering 40

Modeling Building and Selection 42

An Overview of Demand Forecasting Modeling Techniques 44

Model Evaluation 46

Model Deployment 48

Forecasting Solution Acceptance 53

Use Case: Demand Forecasting 54

Conclusion 58

Chapter 3 Time Series Data Preparation 61

Python for Time Series Data 62

Common Data Preparation Operations for Time Series 65

Time stamps vs. Periods 66

Converting to Timestamps 69

Providing a Format Argument 70

Indexing 71

Time/Date Components 76

Frequency Conversion 78

Time Series Exploration and Understanding 79

How to Get Started with Time Series Data Analysis 79

Data Cleaning of Missing Values in the Time Series 84

Time Series Data Normalization and Standardization 86

Time Series Feature Engineering 89

Date Time Features 90

Lag Features and Window Features 92

Rolling Window Statistics 95

Expanding Window Statistics 97

Conclusion 98

Chapter 4 Introduction to Autoregressive and Automated Methods for Time Series Forecasting 101

Autoregression 102

Moving Average 119

Autoregressive Moving Average 120

Autoregressive Integrated Moving Average 122

Automated Machine Learning 129

Conclusion 136

Chapter 5 Introduction to Neural Networks for Time Series Forecasting 137

Reasons to Add Deep Learning to Your Time Series Toolkit 138

Deep Learning Neural Networks Are Capable of Automatically Learning and Extracting Features from Raw and Imperfect Data 140

Deep Learning Supports Multiple Inputs and Outputs 142

Recurrent Neural Networks Are Good at Extracting Patterns from Input Data 143

Recurrent Neural Networks for Time Series Forecasting 144

Recurrent Neural Networks 145

Long Short-Term Memory 147

Gated Recurrent Unit 148

How to Prepare Time Series Data for LSTMs and GRUs 150

How to Develop GRUs and LSTMs for Time Series Forecasting 154

Keras 155

TensorFlow 156

Univariate Models 156

Multivariate Models 160

Conclusion 164

Chapter 6 Model Deployment for Time Series Forecasting 167

Experimental Set Up and Introduction to Azure Machine Learning SDK for Python 168

Workspace 169

Experiment 169

Run 169

Model 170

Compute Target, RunConfiguration, and ScriptRun Config 171

Image and Webservice 172

Machine Learning Model Deployment 173

How to Select the Right Tools to Succeed with Model Deployment 175

Solution Architecture for Time Series Forecasting with Deployment Examples 177

Train and Deploy an ARIMA Model 179

Configure the Workspace 182

Create an Experiment 183

Create or Attach a Compute Cluster 184

Upload the Data to Azure 184

Create an Estimator 188

Submit the Job to the Remote Cluster 188

Register the Model 189

Deployment 189

Define Your Entry Script and Dependencies 190

Automatic Schema Generation 191

Conclusion 196

References 197

Index 199