Amazon SageMaker Feature Store
Learn how to manage and serve machine learning features with Amazon SageMaker Feature Store, ensuring consistency and scalability for model training and inference.
Amazon SageMaker Feature Store is a fully managed, purpose-built service with persisted feature storage for ML models from creation through production. Data features are the components upon which both the training of and the decisions made by ML models are based. For example, a recommendation system could feature user profiles, activity, and product ratings. These are vital characteristics of the model, and while most of them can be addressed by engineering at the model level, managing these across multiple ML applications is also important for defining successful outcomes.
SageMaker Feature Store can potentially ingest features from various data sources, such as real-time streaming and batch data, application/service logs, clickstreams, sensor data, and tabular data from Amazon S3, Amazon Redshift, Snowflake, and Databricks Delta Lake. This enables us to combine various data feeds into our ML processes for use in the training and prediction phases.
A major data consistency problem in machine learning workflows is the use of features in training and during inference. SageMaker Feature Store resolves this by providing an offline store for model building and an online store for real-time inference. This keeps the features used to ...