book

Learning AutoML

Name: Learning AutoML
Author: Kerem Tomak
ISBN: 9798341643185

by Kerem Tomak

April 2026

Intermediate to advanced

586 pages

16h 26m

English

O'Reilly Media, Inc.

Read now

Unlock full access

Foreword by Thomas H. Davenport
Foreword by Gregory Wheeler
Preface
Why I Wrote This BookWho Should Read This BookHow This Book Is OrganizedPart I: Foundations of AutoMLPart II: Core AutoML TechniquesPart III: AutoML for Different Data TypesPart IV: Production and MLOpsPart V: Case StudiesReading Strategies for Different GoalsHands-On Learning ApproachConventions Used in This BookUsing Code ExamplesO’Reilly Online LearningHow to Contact UsAcknowledgments
I. Foundations of AutoML
1. What Is Automated Machine Learning?
The Growing Demand for Machine Learning SolutionsAddressing the Data Science Talent GapDemocratizing AI DevelopmentAutoML in the Machine Learning LandscapeOpen Source AutoML LibrariesEnterprise AutoML PlatformsComparison of Leading FrameworksWho Should Use AutoML?AutoML Across Industries: Transforming Business ProcessesFinanceHealthcare and Life SciencesRetail and EcommerceManufacturingOther SectorsThe Tiered Use Case ModelOvercoming Hurdles: Persistent Challenges in AutoMLInterpretability (the “Black Box” Problem)Need for Customization Versus AutomationData Quality Dependency and RobustnessComputational Costs and Resource IntensityAddressing Bias and FairnessScalability and EfficiencyThe Horizon: Future Trends Shaping AutoMLSynergy with Large Language Models (LLMs) and Foundation ModelsNext-Generation Neural Architecture Search (NAS)Maturity of Multimodal Explainable AI (MXAI)Continued Democratization via Low-Code/No-CodeExpansion to Edge Computing and Federated LearningSummary
2. The Rise and Current State of AutoML
Early Automation (Pre-2010): Laying the GroundworkFeature SelectionHyperparameter SearchMeta-Learning ResearchLimitations of Early ApproachesFirst Generation (2010–2015): Solving the CASH ProblemAuto-WEKA (2013)Hyperopt (2013)SMAC (Sequential Model-Based Algorithm Configuration)First Generation Impact and LegacySecond Generation (2015–2020): Solving the Usability and Enterprise ProblemAuto-sklearn (2015)PyCaret (2020)TPOT (Tree-Based Pipeline Optimization Tool) (2016)H2O AutoMLGoogle Cloud AutoML (Now Part of Vertex AI)Second Generation Impact and LegacyThird Generation (2020–Present): Solving the Multimodal and MLOps ProblemAutoGluon (Amazon)Google Vertex AIMLJAR and AWS SageMaker AutopilotThird Generation Key CapabilitiesThe Emergence of LLM-Assisted AutoMLSummary
3. Understanding the AutoML Pipeline
The Architecture of Automated Machine LearningData PreprocessingData Quality Assessment and CleaningMissing Data StrategiesData Validation and Integrity ChecksFeature EngineeringMultilevel Feature GenerationDomain-Specific Feature EngineeringFeature Selection and PruningRepresentation Learning IntegrationHyperparameter OptimizationAdvanced Search StrategiesMultifidelity OptimizationConfiguration Space DesignBudget-Aware OptimizationNeural Architecture SearchSearch Space EngineeringEfficiency TechniquesHardware-Aware Architecture SearchArchitecture Transfer and Meta-LearningModel Selection, Ensembling, and StackingDiversity-Driven Ensemble ConstructionAdvanced Stacking TechniquesDynamic Ensemble SelectionResource-Aware Ensemble OptimizationModel Deployment and MonitoringProduction Readiness ConsiderationsScalability and Performance OptimizationModel Monitoring and MaintenanceInterpretability and ExplainabilityPipeline Integration and OptimizationCross-Stage Optimization StrategiesResource Allocation and ManagementFeedback Mechanisms and Continuous LearningChallenges and Future DirectionsScalability and EfficiencyRobustness and ReliabilityDemocratization and AccessibilityThe Double-Edged Sword of DemocratizationSummary
II. Core AutoML Techniques
4. Automated Data Preprocessing and Feature Engineering
Working Dataset: RetailMart Ecommerce PlatformIntelligent Data Profiling and Quality AssessmentSmart Data Type Handling and TransformationDateTime Feature ExtractionText Preprocessing PipelinesAutomated Feature EngineeringTraditional Feature Engineering AutomationAdvanced Feature Learning TechniquesIntelligent Feature Selection and Dimensionality ManagementPreprocessing Complex and Multimodal DataProduction-Ready Preprocessing PipelinesSummary
5. Hyperparameter Optimization
The Challenge of Hyperparameter OptimizationThe Computational Cost ChallengeThe Sensitivity ProblemReal-World ImpactGrid Search Versus Random Search: Building the FoundationGrid Search: Systematic but LimitedRandom Search: A Surprisingly Effective AlternativeA Practical ComparisonModern ImplementationWhen to Use Each ApproachLimitations of Both ApproachesBayesian Optimization: Learning from ExperienceThe Core InsightSurrogate ModelsAcquisition FunctionsReal-World Success StoriesModern Tools and ImplementationAWS SageMaker Automatic Model TuningAdvanced TechniquesPractical ConsiderationsLimitations and ChallengesEarly Stopping and Scheduling: Working Smarter, Not HarderThe Core InsightSuccessive Halving: A Tournament ApproachHyperband: Automating the Resource AllocationAsynchronous Successive Halving (ASHA)Population-Based Training: Evolution During TrainingLayer Freezing: A Novel Fidelity DimensionPractical ImplementationReal-World ResultsCombining with Bayesian OptimizationWhen Early Stopping Works BestA Cautionary NoteMultifidelity Optimization: Beyond Simple Early StoppingThe Multifidelity ParadigmAdvanced Multifidelity MethodsPractical ImplementationCase Study: Personal Investment Portfolio Optimization with Multifidelity HPOBackground and Problem DefinitionDataset and FeaturesMultifidelity Strategy ImplementationImplementationResource Management and ResultsKey Insights and Practical ConsiderationsEarly Stopping EffectivenessModel Type Performance PatternsResource Allocation StrategyProduction Deployment ConsiderationsLessons for Individual PractitionersWhen to Use Multifidelity OptimizationSummary

6. Neural Architecture Search (NAS)
Understanding Neural Architecture SearchThe Three Pillars of NASSearch Space Design: Defining the BoundariesThe Art of ConstraintTypes of Search SpacesTask-Specific ConsiderationsEmerging Specialized Search SpacesMulti-Objective Search SpacesBalancing Efficiency and DiscoveryThe NAS-Bench RevolutionSearch Strategies: Finding Needles in HaystacksThe Evolution of Search StrategiesChoosing Your Search Strategy: A Practical PerspectiveReinforcement Learning: The Original ApproachEvolutionary Algorithms: Nature-Inspired SearchDifferentiable NAS: The Game ChangerGradient-Based Methods and Advanced TechniquesHybrid Approaches: Best of Both WorldsChoosing the Right StrategyPerformance Estimation: The Efficiency ImperativeBridging the Evaluation GapThe Training BottleneckMultifidelity Evaluation: Training Less, Learning MoreOne-Shot Architecture Search: Train Once, Evaluate ManyLearning Curve ExtrapolationZero-Cost Proxies: Instant Architecture EvaluationSurrogate Models: Learning to Predict PerformanceCombining Approaches for Maximum EfficiencyEfficient NAS: Making It PracticalThe Efficiency RevolutionBuilding Production-Ready NAS SystemsWeight Sharing: The Foundation of Efficient NASOnce-For-All Networks: Decoupling Training and DeploymentProgressive Search StrategiesHardware-Aware OptimizationProduction Deployment Tools for NAS-Discovered ArchitecturesZero-Cost Proxies for Rapid FilteringPractical Implementation GuidelinesPractical Applications and ToolsNAS in the Real World: Integration and DeploymentAutoKeras: Simplicity FirstNNI: Enterprise-Grade NASRay Tune + Optuna: Flexible and PowerfulIndustry Success StoriesFrom Notebook to Production: Next StepsSummary
III. AutoML for Different Data Types
7. AutoGluon for Tabular Data
Setting Up AutoGluon and EnvironmentInstallation OptionsPlatform-Specific NotesSetting Up Your Development EnvironmentPerformance ConsiderationsCloud Environment RecommendationsChoosing the Right AutoML Framework for Tabular DataTabularPredictor BasicsLoading and Exploring DataBasic Model TrainingUnderstanding TabularPredictor OutputDifferent Prediction MethodsBinary and Multiclass ClassificationBinary Classification in DetailMulticlass ClassificationRegression TasksRegression Versus Classification DifferencesInterpreting Regression PerformanceCustomizing Basic BehaviorAutoGluon’s Automatic Data ProcessingAutomatic Feature Type DetectionMissing Value HandlingCategorical EncodingAdvanced CustomizationCustom HyperparametersAdvanced Ensemble ConfigurationFeature Engineering ControlTraining Process OptimizationModel Interpretability and DebuggingInterpretability ToolsHandling Special Data ScenariosWhen to Use Advanced CustomizationProject: Titanic Survival PredictionProject Overview and Business ContextData Exploration and UnderstandingBaseline AutoGluon ModelCustom Feature Engineering for TitanicModel Interpretation for TitanicPerformance Evaluation and ComparisonModel Deployment PreparationProject Summary and Business ImpactExtending This ProjectData Pipeline ConsistencyMonitoring and Maintaining Models in ProductionMonitoring PracticesMonitoring Tools for AutoGluonSummary
8. AutoML for Text and Natural Language Processing
AutoGluon’s MultiModalPredictor for Text ProcessingWhy MultiModalPredictor?Underlying Model ArchitecturesReal-World PerformanceBuilding Text Classification ModelsYour First Text Classification ModelUnderstanding Model SelectionHyperparameter Optimization GuidelinesAdvanced Text Processing CapabilitiesBeyond Classification: Advanced NLP TasksThe Transformer Revolution and BeyondDomain-Specific ConsiderationsModel Selection for Different Use CasesMaximum Accuracy ApplicationsBalanced ApplicationsHigh-Throughput ApplicationsReal-World Applications and PerformanceIndustry Case StudiesPerformance InsightsProduction Deployment ConsiderationsModel Selection for Deployment ScenariosManaged Services Versus Custom ModelsDeploying Custom Models with SageMakerMonitoring and MaintenancePerformance and Operational MonitoringData Drift DetectionRetraining and Continuous ImprovementPractical Project: News Article ClassificationSummary
9. Time Series Forecasting with AutoGluon
Understanding the Time Series ChallengeGetting Started with TimeSeriesPredictorFoundation Models and Zero-Shot ForecastingThe Chronos-Bolt ArchitectureReal-World Impact of Zero-Shot ForecastingHandling Complex Multiseries ScenariosAdvanced Capabilities: Covariate RegressorsImplementing Covariate RegressorsBusiness Impact of Covariate IntegrationModel Selection and Hyperparameter OptimizationThe Model ZooPreset ConfigurationsCustom Hyperparameter ConfigurationEvaluation and Validation StrategiesBacktesting and Time-Aware ValidationBusiness-Relevant MetricsProduction Deployment and Cloud IntegrationAWS Deployment OptionsModel Updating and MonitoringPractical Project: Retail Demand ForecastingData Preparation and ExplorationModel Training with Advanced FeaturesBusiness Impact AnalysisFuture Directions and Emerging CapabilitiesSummary
10. Computer Vision with AutoGluon
Understanding AutoGluon’s Computer Vision CapabilitiesChoosing Between Custom Models and Managed ServicesBuilding Training Datasets with SageMaker Ground TruthThe MultiModalPredictor AdvantageFoundation Models IntegrationModern Computer Vision ArchitecturesTask Categories and ApplicationsSetting Up AutoGluon for Computer VisionInstallation and Environment SetupHardware ConsiderationsVerification and Basic SetupImage Classification with MultiModalPredictorYour First Image Classification ModelUnderstanding Data Formats and PreprocessingModel Architecture Selection and PresetsAdvanced Classification TechniquesObject Detection with AutoGluonUnderstanding Object DetectionBasic Object Detection SetupEnhanced Object Detection CapabilitiesAdvanced Object Detection ApplicationsMultimodal Computer Vision ApplicationsCombining Images with Tabular DataImage and Text IntegrationReal-World Computer Vision Project: Automated Ecommerce Product ClassificationProject: Automated Ecommerce Product ClassificationData Preparation and ExplorationBuilding the Multimodal Classification SystemPerformance Analysis and Model InterpretabilityIntegration with Ecommerce SystemsPerformance Optimization and Best PracticesHardware Optimization StrategiesModel Monitoring and MaintenanceProduction Deployment ConsiderationsModel Versioning and UpdatesSageMaker Endpoint DeploymentSageMaker Serverless Inference for Cost-Effective DeploymentAWS Panorama for Edge DeploymentScalable Batch Processing ServiceSummary
IV. Production and MLOps
11. Workflow Integration with MLOps Tools
Understanding the AutoML-MLOps Integration LandscapeThe Scale ChallengeThe Reproducibility ImperativeExperiment Tracking and Model ManagementHierarchical Experiment OrganizationArtifact Management StrategiesWorkflow Orchestration with KubeflowDesigning AutoML-Aware PipelinesResource Management and OptimizationProduction Deployment PatternsAutomated Validation and Quality AssuranceDynamic Serving InfrastructureOperational Monitoring and MaintenanceMonitoring and GovernanceAutoML-Specific Monitoring RequirementsGovernance and Compliance FrameworksIntegration Challenges and SolutionsThe Artifact Explosion ChallengeEnsuring Reproducibility in Automated SystemsBridging Technical and Business DomainsBest Practices and Implementation GuidelinesBuilding Progressive CapabilityAlignment and Expectation ManagementRisk Management and Parallel SystemsObservability as a FoundationOrganizational Learning and AdaptationSummary
12. Data Pipeline Automation with Apache Airflow
Understanding Data Pipeline Requirements for AutoMLAirflow Architecture for Machine Learning WorkflowsCore ComponentsKey Airflow TerminologyDesigning DAGs for AutoML Data IngestionPractical Example: Complete AutoML Data Ingestion DAGDAG Initialization and ConfigurationUnderstanding catchup BehaviorDynamic Task Mapping for Parallel ProcessingFeature Engineering Pipelines and Feature StoresHandling Late-Arriving DataData Contracts and Schema EvolutionMonitoring and Data Quality GatesScaling Airflow for Enterprise AutoMLOperational Excellence and Best PracticesSummary
13. Deployment and Continuous Delivery for AutoML
The Unique Challenges of AutoML DeploymentContinuous Integration for Machine LearningShadow Deployment ValidationContinuous Deployment PipelinesTesting Strategies for Automated ModelsContract TestingProperty-Based TestingMetamorphic TestingAdversarial TestingModel Packaging and ContainerizationPractical Example: Deploying the Adult Income Prediction ModelModel Serving InfrastructureMonitoring and Observability in ProductionPrometheus–Grafana Monitoring StackDrift Detection with EvidentlySecurity and Compliance ConsiderationsInput Sanitization and DoS PreventionAdversarial Attack DefensesContinuous Learning and Feedback LoopsSummary
V. Case Studies
14. Case Study 1: Financial Services—Real-Time Fraud Detection at GlobalBank
Business Problem and ContextSuccess CriteriaData Pipeline and PreparationData Pipeline ArchitectureProduction Data Pipeline ConsiderationsFeature Engineering1. Temporal Features: Fraud Has a Schedule2. Velocity Features: Fraudsters Move Fast3. Behavioral Deviation: Detecting the Unusual4. Merchant Risk Scoring5. Device TrustFeature Impact SummaryModel Development with AutoGluonSample Weighting for Cost-Sensitive LearningAutoGluon ConfigurationWhy PR-AUC Instead of ROC-AUC?Model Training ResultsModel Evaluation and InterpretabilityFinding the Optimal ThresholdModel Interpretability with SHAPDeployment ArchitectureFastAPI Inference ServiceGraceful Degradation StrategyMonitoring and MaintenanceDrift Detection with PSIAutomated Retraining PipelineA/B Testing for Model UpdatesOutcomes and Lessons LearnedPerformance MetricsBusiness ImpactKey LessonsThe Right Division of LaborSummary
15. Case Study 2: Retail—Omnichannel Demand Forecasting
Business Problem and ContextThe Scale ChallengeThe Wake-Up CallProject ObjectivesData Challenges: Multisource IntegrationPoint-of-Sale DataEcommerce DataInventory DataMarketing and Promotions DataWeather DataExternal SignalsData Pipeline ArchitectureKey Data DecisionsFeature Engineering: Capturing Demand DriversBusiness-Aligned MetricsTemporal Features (Baseline)Omnichannel Behavioral Features (High Impact)Weather-Driven Demand (Category-Specific)Promotional Features (Complex Interactions)Event-Driven DemandSKU-Specific AttributesModel Development: AutoGluon for Time Series at ScaleThe AutoML ApproachWhy Tabular AutoML for Time Series?Training Strategy: Time-Based SplitsMultihorizon ForecastingAutoGluon ConfigurationKey Configuration DecisionsHandling Data Sparsity (Long-Tail SKUs)Training InfrastructureEvaluation: Business Metrics over Model MetricsModel Performance (MAPE by Forecast Horizon)Weighted Average MAPE (Business-Aligned)Business Impact MetricsCategory-Specific PerformancePromotional Forecast AccuracyForecast Bias AnalysisDeployment: Production Forecasting PipelinePipeline ArchitectureTechnology StackForecast ServingMonitoring: Keeping Forecasts AccurateDrift DetectionBusiness Outcomes and Lessons LearnedQuantified Business Impact (12 Months Postlaunch)Unexpected BenefitsCritical Success FactorsWhat We’d Do DifferentlyLessons for Your Demand Forecasting ProjectWhen AutoML Excels for Demand ForecastingSummary
16. Case Study 3: Healthcare—Patient Readmission Prediction
The Business ChallengeThe Hardest Constraint: FairnessThe Current StateProject ObjectivesData Challenges and HIPAA ComplianceData Sources and IntegrationMissing Data PatternsData Quality IssuesFeature Engineering: Structured and Unstructured DataCategory 1: Demographics and Social Determinants (42 Features)Category 2: Clinical Complexity and Comorbidities (68 features)Category 3: Utilization History (53 features)Category 4: Current Encounter Features (87 features)Category 5: Clinical Notes Embeddings (64 features)Category 6: Temporal and Interaction Features (33 features)Model Development: Fairness-Aware AutoMLThe Fairness ChallengeFairness Metrics DefinedBaseline Model: Standard AutoGluon (No Fairness Constraints)Approach 1: Remove Protected AttributesApproach 2: Adversarial DebiasingApproach 3 (Final): Fairness-Aware Ensemble with ReweightingFinal Model ConfigurationEvaluation: Performance + Fairness MetricsModel Performance (Overall)Fairness Metrics by RaceFairness by Age GroupFeature Importance (Top 20 by SHAP)Business MetricsDeployment: Clinical Workflow IntegrationReal-Time Prediction ArchitectureEHR Integration (Epic)Clinical Decision Support AlertCare Management WorkflowClinician Training and Change ManagementInterpretability for CliniciansRegulatory ConsiderationsMonitoring: Drift and FairnessThree-Layer Monitoring StrategyRetraining ScheduleBusiness Outcomes and Lessons LearnedClinical OutcomesFairness in PracticeUnexpected BenefitsCritical Success FactorsWhat We’d Do DifferentlyLessons for Your Readmission ProjectWhen AutoML Excels for HealthcareThe Production AutoML Blueprint: A Grand SynthesisThe Universal Patterns of Production AutoMLThe Production Readiness ChecklistFinal ThoughtsSummary
Epilogue: The Quiet AutoML Revolution
The Original Promise, DeliveredThe Assumptions We Built OnThe Foundation Model DisruptionFrom Automation to AutonomyWhat Agentic AutoML Actually Looks LikeThe New Skills GapWhy Everything You Learned Still MattersThe Governance Question Nobody’s AnsweringThe Road AheadA Final Word
Index
About the Author

Overview

Learning AutoML is your practical guide to applying automated machine learning in real-world environments. Whether you're a data scientist, ML engineer, or AI researcher, this book helps you move beyond experimentation to build and deploy high-performing models with less manual tuning and more automation. Using AutoGluon as a primary toolkit, you'll learn how to build, evaluate, and deploy AutoML models that reduce complexity and accelerate innovation.

Author Kerem Tomak shares insights on how to integrate models into end-to-end deployment workflows using popular tools like Kubeflow, MLflow, and Airflow, while exploring cross-platform approaches with Vertex AI, SageMaker Autopilot, Azure AutoML, Auto-sklearn, and H2O.ai. Real-world case studies highlight applications across finance, healthcare, and retail, while chapters on ethics, governance, and agentic AI help future-proof your knowledge.

Build AutoML pipelines for tabular, text, image, and time series data
Deploy models with fast, scalable workflows using MLOps best practices
Compare and navigate today's leading AutoML platforms
Interpret model results and make informed decisions with explainability tools
Explore how AutoML leads into next-gen agentic AI systems

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9798341643178Errata Page

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills