Amazon SageMaker Feature Store

 Amazon SageMaker Feature Store is a managed feature store from AWS that gives machine learning teams a central place to create, store, and serve features for training and inference. It maintains an online store for low-latency lookups during real-time prediction and an offline store in Amazon S3 for building training datasets, and it keeps the two in sync. Point-in-time queries reconstruct feature values as they existed at a past moment, which prevents data leakage when assembling training data, and the store integrates with the wider SageMaker platform.

Key Capabilities:

• An online store for low-latency feature lookups during inference

• An offline store in Amazon S3 for assembling training datasets

• Point-in-time queries that prevent leakage in training data

• Feature groups that organize and version related features

• Integration with SageMaker pipelines and the AWS ecosystem

• Access control and encryption through AWS identity and KMS