book

Implementing MLOps in the Enterprise

by Yaron Haviv, Noah Gift

December 2023

Intermediate to advanced

377 pages

9h 21m

English

O'Reilly Media, Inc.

Book available

Read now

Unlock full access

Who This Book Is ForNavigating This BookConventions Used in This BookUsing Code ExamplesO’Reilly Online LearningHow to Contact UsAcknowledgmentsYaronNoah
What Is MLOps?MLOps in the EnterpriseUnderstanding ROI in Enterprise SolutionsUnderstanding Risk and Uncertainty in the EnterpriseMLOps Versus DevOpsWhat Isn’t MLOps?Mainstream Definitions of MLOpsWhat Is ML Engineering?MLOps and Business IncentivesMLOps in the CloudKey Cloud Development EnvironmentsThe Key Players in Cloud ComputingMLOps On-PremisesMLOps in Hybrid EnvironmentsEnterprise MLOps StrategyConclusionCritical Thinking Discussion QuestionsExercises
Getting StartedChoose Your AlgorithmDesign Your PipelinesData Collection and PreparationData Storage and IngestionData Exploration and PreparationData LabelingFeature StoresModel Development and TrainingWriting and Maintaining Production ML CodeTracking and Comparing Experiment ResultsDistributed Training and Hyperparameter OptimizationBuilding and Testing Models for ProductionDeployment (and Online ML Services)From Model Endpoints to Application PipelinesOnline Data PreparationContinuous Model and Data MonitoringMonitoring Data and Concept DriftMonitoring Model Performance and AccuracyThe Strategy of Pretrained ModelsBuilding an End-to-End Hugging Face ApplicationFlow Automation (CI/CD for ML)ConclusionCritical Thinking Discussion QuestionsExercises
Identifying the Business Use Case and GoalsFinding the AI Use CaseDefining Goals and Evaluating the ROIHow to Build a Successful ML ProjectApproving and Prototyping the ProjectScaling and Productizing ProjectsProject Structure and LifecycleML Project Example from A to ZExploratory Data AnalysisData and Model Pipeline DevelopmentApplication Pipeline DevelopmentScaling and Productizing the ProjectCI/CD and Continuous OperationsConclusionCritical Thinking Discussion QuestionsExercises
Data Versioning and LineageHow It WorksCommon ML Data Versioning ToolsData Preparation and Analysis at ScaleStructured and Unstructured Data TransformationsDistributed Data Processing ArchitecturesInteractive Data ProcessingBatch Data ProcessingStream ProcessingStream Processing FrameworksFeature StoresFeature Store Architecture and UsageIngestion and Transformation ServiceFeature StorageFeature Retrieval (for Training and Serving)Feature Stores Solutions and Usage ExampleUsing Feast Feature StoreUsing MLRun Feature StoreConclusionCritical Thinking Discussion QuestionsExercises
AutoMLRunning, Tracking, and Comparing ML JobsExperiment TrackingSaving Essential Metadata with the Model ArtifactsComparing ML Jobs: An Example with MLflowHyperparameter TuningAuto-LoggingMLOps Automation: AutoMLOpsExample: Running and Tracking ML Jobs Using Azure DatabricksHandling Training at ScaleBuilding and Running Multi-Stage WorkflowsManaging Computation Resources EfficientlyConclusionCritical Thinking Discussion QuestionsExercises
Model Registry and ManagementSolution ExamplesSageMaker ExampleMLflow ExampleMLRun ExampleModel ServingAmazon SageMakerSeldon CoreMLRun ServingAdvanced Serving and Application PipelinesImplementing Scalable Application PipelinesModel Routing and EnsemblesModel Optimization and ONNXData and Model MonitoringIntegrated Model Monitoring SolutionsStandalone Model Monitoring SolutionsModel RetrainingWhen to Retrain Your ModelsStrategies for Data RetrainingModel Retraining in the MLOps PipelineDeployment StrategiesMeasuring the Business ImpactConclusionCritical Thinking Discussion QuestionsExercises
Exploratory Data AnalysisInteractive Data PreparationPreparing the Credit Transaction DatasetPreparing the User Events (Activities) DatasetExtracting Labels and Training a ModelData Ingestion and Preparation Using a Feature StoreBuilding the Credit Transactions Data Pipeline (Feature Set)Building the User Events Data Pipeline (FeatureSet)Building the Target Labels Data Pipeline (FeatureSet)Ingesting Data into the Feature StoreModel Training and Validation PipelineCreating and Evaluating a Feature VectorBuilding and Running an Automated Training and Validation PipelineReal-Time Application PipelineDefining a Custom Model Serving ClassBuilding an Application Pipeline with Enrichment and EnsembleTesting the Application Pipeline LocallyDeploying and Testing the Real-Time Application PipelineModel MonitoringCI/CD and Continuous OperationsConclusionCritical Thinking Discussion QuestionsExercises
Distributed Deep LearningHorovodRayData Gathering, Labeling, and Monitoring in DLData Labeling Pitfalls to AvoidData Labeling Best PracticesData Labeling SolutionsUsing Foundation Models as LabelersMonitoring DL Models with Unstructured DataBuild Versus Buy Deep Learning ModelsFoundation Models, Generative AI, LLMsRisks and Challenges with Generative AIMLOps Pipelines for Efficiently Using and Customizing LLMsApplication Example: Fine-Tuning an LLM ModelConclusionCritical Thinking Discussion QuestionsExercises
ML Problem Framing with Time SeriesNavigating Time Series Analysis with AWSDiving into Time Series with DeepAR+Time Series with the GCP BigQuery and SQLBuild Versus Buy for MLOps NLP ProblemsBuild Versus Buy: The Hugging Face ApproachExploring Natural Language Processing with AWSExploring NLP with OpenAIVideo Analysis, Image Classification, and Generative AIImage Classification Techniques with CreateMLComposite AIGetting Started with Serverless for Composite AIUse Cases of Composite AI with ServerlessConclusionCritical Thinking Discussion QuestionsExercises

The Case for Rust for MLOpsLeveling Up with Rust, GitHub Copilot, and CodespacesIn the Beginning Was the Command LineGetting Started with Rust for MLOpsUsing PyTorch and Hugging Face with RustUsing Rust to Build Tools for MLOpsBuilding Containerized Rust Command-Line ToolsGPU PyTorch WorkflowsUsing TensorFlow RustDoing k-means Clustering with RustFinal Notes on RustRuff Linterrust-new-project-templateConclusionCritical Thinking Discussion QuestionsExercises

Content preview from Implementing MLOps in the Enterprise

Chapter 1. MLOps: What Is It and Why Do We Need It?

At the root of inefficient systems is an interconnected web of incorrect decisions that compound over time. It is tempting to look for a silver bullet fix to a system that doesn’t perform well, but that strategy rarely, if ever, pays off. Consider the human body; there is no shortage of quick fixes sold to make you healthy, but the solution to health longevity requires a systematic approach.¹

Similarly, there is no shortage of advice on “getting rich quick.” Here again, the data conflicts with what we want to hear. In Don’t Trust Your Gut (HarperCollins, 2022), Seth Stephens-Davidowitz shows that 84% of the top 0.1% of earners receive at least some money from owning a business. Further, the average age of a business founder is about 42, and some of the most successful companies are real estate or automobile dealerships. These are hardly get-rich-quick schemes but businesses that require significant skill, expertise, and wisdom through life experience.

Cities are another example of complex systems that don’t have silver bullet fixes. WalletHub created a list of best-run cities in America with San Francisco ranked 149 out of 150 despite having many theoretical advantages over other cities, like beautiful weather, being home to the top tech companies in the world, and a 2022-2023 budget of $14 billion for a population of 842,000 people. The budget is similar to the entire country of Panama, with a population of 4.4 million people. ...