Exam AWS Certified Machine Learning - Specialty topic 1 question 257 discussion

Answer is A - "SageMaker Feature Store consists of an online and an offline mode for managing features. The online store is used for low-latency real-time inference use cases. The offline store is primarily used for batch predictions and model training." https://aws.amazon.com/blogs/machine-learning/speed-ml-development-using-sagemaker-feature-store-and-apache-iceberg-offline-store-compaction/

Best Choice: B. Amazon SageMaker Feature Store This option offers a centralized and efficient solution that meets all the requirements: Feature Storage: SageMaker Feature Store acts as a single repository for features used in both offline training and online inference. Online Store: Creating an online store within Feature Store eliminates the need for a separate S3 bucket for inference features, simplifying management. Feature History Tracking: Feature Store automatically tracks feature lineage and versions, allowing you to see changes and roll back if needed. Data Science Team Access: IAM roles can be created to grant data scientists access to search and explore features within Feature Groups in the store.

A meets all requirements, and looks like the easiest to setup

A. YES - online store will be faster for inference, offline store cheaper for batch B. NO - online store for offline will be too expensive C. NO - want to use Feature store D. NO - want to use Feature store

Amazon SageMaker Feature Store is a managed service that makes it easy to store and manage features for machine learning models. It provides a scalable and reliable way to store features, and it supports both online inference and offline model training. Creating separate online and offline stores in SageMaker Feature Store will allow the music streaming company to optimize the storage and performance of their features for each use case. The online store can be configured to be highly available and performant, while the offline store can be configured to be cost-effective and scalable.