Amazon SageMaker Feature Store: Store, discover, & share features for ML apps
summary
Amazon SageMaker Feature Store is a managed feature repository that stores features for online and offline store use, supports streaming and batch ingest (custom Spark container via SageMaker Data Wrangler), and enforces schema validation. It promises low millisecond latency lookups, <15 minute read-after-write consistency to the offline parquet store, and time‑travel reconstruction using event time + write time. Demo shows Athena queries to build training sets, XGBoost training and deployment, and a live GET-record lookup producing a 0.8 fraud probability. Feature groups hold up to 2,500 features and integrate with Glue, S3, KMS, and IAM.
AI-written summary · about
chapters (18)
- 0:00Introduction to SageMaker Feature Store
- 1:05What are features and why they matter
- 2:53Challenges in scaling ML features
- 5:05How Feature Store solves the problem
- 7:12Core capabilities of SageMaker Feature Store
- 8:16Feature groups and organization
- 11:07Consistency between online and offline stores
- 12:34Data ingestion methods
- 13:58Security and access control
- 15:03Demo: Creating feature groups
- 20:40Demo: Discovering features
- 21:47Demo: Ingesting and training data
- 23:13Demo: Serving features for predictions
- 24:39Demo: Exploring data with Data Wrangler
- 25:22Time travel with historical features
- 27:52Customer success stories
- 28:54Pricing and getting started
- 29:35Q&A and closing
related talks
44mSimplify Data Preparation for Machine Learning
58mPrepare data and model features for ML with ease, speed, and accuracy
32mA feature Store for Machine Learning
44mExploratory data analysis and automated feature engineering
30mHow Feature Stores Enable Operational ML
35mPanel 03- How to make the Feature Store become a hidden part of the productionalization of ML models
20mNavigating the Skies of Data: A Flight Through Feature Stores
34mPrepare data for machine learning with ease, speed, and accuracy