AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsEasy

A data science team is developing a fraud detection model. They need a centralized, version-controlled repository to store, share, and manage features used across multiple models and teams, ensuring consistency between training and inference. The solution must support both online (low-latency) and offline (batch) access patterns. Which AWS service is best suited for this requirement?

  1. AAmazon SageMaker Feature Store
  2. BAmazon S3
  3. CAWS Glue Data Catalog
  4. DAmazon DynamoDB
Show answer & explanation

Correct answer: A. Amazon SageMaker Feature Store

Amazon SageMaker Feature Store is a purpose-built service for storing, updating, and serving machine learning features for training and inference. It supports both online (low-latency) and offline (batch processing) access, ensuring consistency and reusability.

Why the other options are wrong

  • B. Amazon S3 can store raw data and feature sets, but it lacks built-in capabilities for feature versioning, online serving, and consistent access patterns for ML.
  • C. AWS Glue Data Catalog organizes metadata for data lakes but doesn't store the feature data itself or provide online serving capabilities for ML.
  • D. Amazon DynamoDB is a key-value and document database that could serve features online, but it's not purpose-built for feature management, versioning, or offline access for ML workflows.

SageMaker Feature Store

A fully managed service in Amazon SageMaker that provides a centralized repository to store, retrieve, and share machine learning features for consistency between training and inference.

  • Supports online (low-latency) and offline (batch) access.
  • Ensures feature consistency for training and inference.
  • Provides feature versioning and discoverability.

Memory trick: Feature Store keeps your ML data organized and ready for action.

More Machine Learning Implementation and Operations questions