Panel 03- How to make the Feature Store become a hidden part of the productionalization of ML models
summary
Panelists argue feature stores must stop being a separate product and become an orchestrator plus metadata layer that integrates with Snowflake, BigQuery, S3/Parquet and existing online stores like Redis rather than replace them. They show patterns: Python APIs that translate to SQL, offline data on S3, online cache vs offline store separation, and pragmatic double implementations in Spark and Flink or Beam for streaming versus batch. The takeaway is to push toward a single feature definition while accepting cost, latency and tooling tradeoffs.
AI-written summary · about
chapters (9)
- 0:00Panel introduction
- 2:06What is a feature store
- 9:09Feature store vs database
- 13:42Feature store integration
- 16:51Data warehouse limitations
- 20:22Online vs offline features
- 23:06Batch vs streaming pipelines
- 28:00Designing for stakeholders
- 32:12SQL vs Python debate
related talks
30mHow Feature Stores Enable Operational ML
20mNavigating the Skies of Data: A Flight Through Feature Stores
15mFeature Stores: Powering Machine Learning Infrastructure with Apache Flink
32mA feature Store for Machine Learning
30mDataEngBytes 2023 - MEL-T2-01 - Mikiko Bazeley
30mAmazon SageMaker Feature Store: Store, discover, & share features for ML apps
19mML feature store a cornerstone in modern data architecture for AI/ML
31mBuilding an open-source ML feature store with Apache Flink