book

Feature Store for Machine Learning

by Jayanth Kumar M J

June 2022

Beginner to intermediate

280 pages

5h 33m

English

Packt Publishing

Read now

Unlock full access

ContributorsAbout the authorAbout the reviewer
Who this book is forWhat this book coversTo get the most out of this bookDownload the example code filesDownload the color imagesConventions usedGet in touchShare Your Thoughts
Technical requirementsThe ML life cycle in practiceProblem statement (plan and create)Data (preparation and cleaning)ModelPackage, release, and monitorAn ideal world versus the real worldReusability and sharingEverything in a notebookThe most time-consuming stages of MLFiguring out the datasetData exploration and feature engineeringModeling to production and monitoringSummary
Importance of features in productionWays to bring features to productionBatch model pipelineOnline model pipelineCommon problems with the approaches used for bringing features to productionRe-inventing the wheelFeature re-calculationFeature discoverability and sharingTraining vs Serving skew Model reproducibilityLow latencyFeature stores to the rescueStandardizing ML with a feature storeFeature store avoids reprocessing dataFeatures are discoverable and sharable with the feature store Serving features at low latency with feature storesPhilosophy behind feature storesSummaryFurther reading
Technical requirementsIntroduction to Feast and installationFeast terminology and definitionsFeast initializationFeast usageRegister feature definitionsBrowsing the feature storeAdding an entity and FeatureViewGenerate training dataLoad features to the online storeFeast behind the scenesData flow in FeastSummaryFurther reading
Technical requirementsCreating Feast resources in AWS Amazon S3 for storing dataAWS Redshift for an offline storeCreating an IAM user to access the resourcesFeast initialization for AWSExploring the ML life cycle with FeastProblem statement (plan and create)Data (preparation and cleaning)Model (feature engineering)SummaryReferences
PrerequisitesTechnical requirementsModel training with the feature storeDee's model training experimentsRam's model training experimentsModel packagingBatch model inference with FeastOnline model inference with FeastSyncing the latest features from the offline to the online storePackaging the online model as a REST endpoint with Feast codeHandling changes to the feature set during developmentStep 1 – Change feature definitionsStep 2 – Add/update schema in the Glue/Lake Formation consoleStep 3 – Update notebooks with the changesSummary Further reading
Technical requirementsSetting up Airflow for orchestrationS3 bucket for Airflow metadataAmazon MWAA environment for orchestrationProductionizing the batch model pipelineProductionizing an online model pipelineOrchestration of a feature engineering jobDeploying the model as a SageMaker endpointBeyond model productionFeature drift monitoring and model retrainingModel reproducibility and prediction issuesA headstart for the next modelChanges to feature definition after production Summary

Technical requirementsThe available feature stores on the marketThe Tecton Feature StoreDatabricks Feature StoreGoogle's Vertex AI Feature StoreThe Hopsworks Feature StoreSageMaker Feature StoreFeature management with SageMaker Feature StoreResources to use SageMakerGenerating featuresDefining the feature groupFeature ingestionGetting records from an online storeQuerying historical data with Amazon AthenaCleaning up a SageMaker feature groupML best practicesData validation at sourceBreaking down ML pipeline and orchestrationTracking data lineage and versioningThe feature repositoryExperiment tracking, model versioning, and the model repositoryFeature and model monitoringMiscellaneousSummary
Technical requirementsInfrastructure setupIntroduction to the problem and the datasetData processing and feature engineeringFeature group definitions and feature ingestionModel trainingModel predictionFeature monitoringModel monitoringSummary
Other Books You May EnjoyPackt is searching for authors like youShare Your Thoughts

Overview

Feature Store for Machine Learning is your comprehensive guide to understanding and utilizing feature stores in your ML pipelines. By exploring core concepts and practical use cases, you'll learn how to streamline data preparation, enable feature sharing and reuse, and reduce redundancy. This book provides actionable insights to improve the efficiency and scalability of your machine learning workflows.

What this Book will help me do

Gain a deep understanding of what feature stores are and their importance in machine learning.
Learn how to effectively curate, store, and organize features for ML pipelines.
Understand strategies to share and reuse machine learning features to reduce redundancy.
Discover techniques for integrating feature stores with both batch and online inference models.
Learn how to design and deploy your own production-ready feature store for real-world applications.

Author(s)

None J, the author of this book, is a seasoned professional in the field of machine learning and data engineering. Known for their practical and hands-on teaching style, None J has a knack for explaining complex technical concepts in an engaging and accessible way. With years of experience working in ML model development, deployment, and operationalization, they bring a deeply practical perspective to the topic of feature stores. None J's focus on empowering data scientists and engineers shines through in their writing.

Who is it for?

This book is tailored for data scientists and ML engineers who have a foundational understanding of machine learning and seek to incorporate feature stores into their workflows. It's perfect for professionals looking to improve collaboration and efficiency within their ML pipelines by leveraging feature stores. If your goals involve optimizing data preparation, enabling feature reuse, and deploying robust machine learning solutions to production, this book is ideal for you.