Machine Learning for Managers

Description

Machine learning can help managers make better predictions, automate complex tasks and improve business operations. Managers who are familiar with machine learning are better placed to navigate the increasingly digital world we live in. There is a view that machine learning is a highly technical subject that can only be understood by specialists. However, many of the ideas that underpin machine learning are straightforward and accessible to anyone with a bit of curiosity. This book is for managers who want to understand what machine learning is about, but who lack a technical background in computer science, statistics or math. The book describes in plain language what machine learning is and how it works. In addition, it explains how to manage machine learning projects within an organization. This book should appeal to anyone that wants to learn more about using machine learning to drive value in real-world organizations. Cover Half Title Title Page Copyright Page Dedication Table of Contents Overview Preface List of Figures List of Tables Author I Understanding Machine Learning 1 Let’s Jump Right in 1.1 What Can We Learn from 34 Lines of Code? 1.2 Fitting ML into the Big Picture 1.3 A Layered Perspective of Machine Learning 1.4 Data, Compute and Methods 1.5 ML Drives Wealth Creation 2 Different Kinds of ML 2.1 An Introduction to the ML Zoo 2.2 Supervised vs Unsupervised ML 2.3 Generative Learning 2.4 Reinforcement Learning 2.5 Online vs Batch Training 2.6 Value-Destroying vs Value-Creating ML 3 Creating ML Models 3.1 Data, Instances and Features 3.2 Targets and Inputs 3.3 Training, Validation and Test Data Sets 3.4 The Machine Learning Recipe 3.4.1 Specify the Problem 3.4.2 Collect the Data 3.4.3 Split the Data 3.4.4 Understand and Explore the Data 3.4.5 Pre-process the Data and Construct Features 3.4.6 Select A Machine Learning Approach 3.4.7 Select Hyper-Parameters 3.4.8 Train the Model 3.4.9 Evaluate the Model on Validation Data 3.4.10 If Validation Performance is Weak 3.4.11 Train the Final Model 3.4.12 Evaluate the Final Model on the Test Data 3.4.13 If Test Performance is Weak 3.4.14 Deploy the Model in Production 3.4.15 Monitor the Model 4 Linear Models 4.1 A Simple Linear Model 4.2 Training Linear Regression Models 4.3 Using Feature Transformations in Linear Models 4.4 Performance Measures for Regression Tasks 4.5 Linear Models with Indicator Variables and Interactions 4.6 Classification with Logistic Regression 4.7 Regularization – Ridge Regression, Lasso and Elastic Net 5 Neural Networks 5.1 A Brief History of Neural Networks 5.2 A Linear Model is a Neural Net (A Very Simple One) 5.3 All You Ever Wanted to Know About Nodes 5.4 More Complex Neural Networks 5.5 Training A Neural Network 5.6 The MNIST Example 5.7 A Peek into the Future – Transformers and Language Models 6 Tree-Based Approaches, Ensembles and Boosting 6.1 The Titanic Example 6.2 Making Predictions with A Tree Model 6.3 Performance Measures for Classification Tasks 6.3.1 Confusion Matrices 6.3.2 Classification Performance Measures 6.3.3 Thresholds and the ROC-AUC Measure 6.4 Ensembles and Random Forests 6.5 Gradient Boosting Machines 7 Dimensionality Reduction and Clustering 7.1 Why Reduce Dimensionality? 7.2 Principal Components Analysis 7.3 Clustering 7.3.1 K-Means Clustering 7.3.2 Hierarchical Agglomerative Clustering (HAC) 8 Unstructured Data 8.1 Images 8.2 Sequences 8.3 Transformers 9 Explainable AI 9.1 Why Do We Need Explainable AI? 9.2 Explaining Models 9.2.1 Intrinsically Interpretable Models 9.2.2 Surrogate Models 9.3 Explaining Predictions II Managing Machine Learning Projects 10 The ML System Lifecycle 10.1 Context 10.2 Identify 10.3 Pilot 10.4 Pipeline 10.5 Development 10.6 Deployment and Monitoring 10.7 The Circle of Life 11 The Big Picture 11.1 Why Getting Things Done is Hard 11.2 Governance Model 11.3 Security and Privacy 11.4 Explain Ability and Fairness 11.5 Laws, Regulations and Compliance 12 Creating Value with ML 12.1 Sources of Value 12.2 The Data-Centric Firm 12.3 The Economics of Platforms 12.4 Outside of Platforms 13 Making the Business Case 13.1 Executive Summary 13.2 Description of the Project 13.3 Project Benefits 13.4 Proof-of-Concept 13.5 Required Resources 13.6 Technical Appendix 14 The ML Pipeline 14.1 Who Needs a Pipeline Anyway? 14.2 The ML Pipeline 15 Development 15.1 A Very Brief Introduction to Software Engineering 15.1.1 Divide and Conquer 15.1.2 Expose Interfaces, Hide Implementations 15.1.3 Implement Incrementally 15.1.4 Use Version Control 15.1.5 Conduct Automated Testing 15.2 Validating the Pipeline 15.2.1 Run-Through Data 15.2.2 Synthetic Data 15.2.3 Trivial Models 15.2.4 Simple Benchmark Models 15.2.5 Current Approach 15.3 Model Development 15.4 Performance vs Value 15.5 Technical Debt 16 Deployment and Monitoring 16.1 Set Up the Production Environment 16.2 Connect the Plumbing 16.3 Test, Test, Test 16.4 Flip the Switch 16.5 Continuous Monitoring 16.6 Final Thoughts Index