Get started with Amazon SageMaker Feature Store - Amazon SageMaker
Feature Store Concepts

Get started with Amazon SageMaker Feature Store

To get started using Amazon SageMaker Feature Store, review the basic concepts, learn how to ingest data for your feature store, and then walk through a Feature Store example. The following sections explain how to create feature groups, ingest data into the groups, and how to manage security for your feature store. 

Topics

Feature Store Concepts

The following list of terms are key to understanding the capabilities of Amazon SageMaker Feature Store: 

  • Feature store – Serves as the single source of truth to store, retrieve, remove, track, share, discover, and control access to features.

  • Feature – A measurable property or characteristic that encapsulates an observed phenomenon. In the Amazon SageMaker Feature Store API, a feature is an attribute of a record. You can define a name and type for every feature stored in Feature Store. Name uniquely identifies a feature within a feature group. Type identifies the datatype for the values of the feature. Supported datatypes are: String, Integral and Fractional. 

  • Feature group – A FeatureGroup is the main Feature Store resource that contains the metadata for all the data stored in Amazon SageMaker Feature Store. A feature group is a logical grouping of features, defined in the feature store, to describe records. A feature group’s definition is composed of a list of feature definitions, a record identifier name, and configurations for its online and offline store. 

  • Feature definition – A FeatureDefinition consists of a name and one of the following data types: an Integral, String or Fractional. A FeatureGroup contains a list of feature definitions. 

  • Record identifier name – Each feature group is defined with a record identifier name. The record identifier name must refer to one of the names of a feature defined in the feature group's feature definitions.

  • Record – A Record is a collection of values for features for a single record identifier value. A combination of record identifier name and a timestamp uniquely identify a record within a feature group. 

  • Event time – a point in time when a new event occurs that corresponds to the creation or update of a record in a feature group. All records in the feature group must have a corresponding Eventtime. It can be used to track changes to a record over time. The online store contains the record corresponding to the last Eventtime for a record identifier name, whereas the offline store contains all historic records. Event time values can either be of a fractional or string type. Fractional values must be UNIX timestamps. Strings must follow the ISO 8601 standard. The following formats are supported yyyy-MM-dd'T'HH:mm:ssZ and yyyy-MM-dd'T'HH:mm:ss.SSSZ where yyyy, MM, and dd represent the year, month, and day respectively and HH, mm, ss, and if applicable, SSS represent the hour, month, second and milliseconds respsectively. T and Z are constants.

  • Online Store – the low latency, high availability cache for a feature group that enables real-time lookup of records. The online store allows quick access to the latest value for a Record via the GetRecord API. A feature group contains an OnlineStoreConfig controlling where the data is stored.

  • Offline store – the OfflineStore, stores historical data in your S3 bucket. It is used when low (sub-second) latency reads are not needed. For example, when you want to store and serve features for exploration, model training, and batch inference. A feature group contains an OfflineStoreConfig controlling where the data is stored.

  • Ingestion – The act of populating feature groups in the feature store.