Data Storage #

Feature Store #

Feature store is a central repository for storing documented, curated and access-controlled features.
Features store enables the team to re-use shared features to avoid redundant work.

Key points:

• managing feature data from a single person to an enterprise
• Scalable and performant access to features for training and serving
• provide consistent and point-in-time correct access
• enable discovery, documentation and insights

Offline feature procesing:

• feature engineering, like feature crossing
• data quality control
• offline storage
• discoverability

Online feature usage:

• low-latency access
• access to pre-computed features (computing aggregations and joins online is very time-consuming)

Data warehoues #

Key features:

• subject-oriented
• integrated (multiple data sources)
• non-volatile (access to history)

Data Lakes #

• Aggregates raw data from multiple sources
• Structured and unsctructured
• Purpose of data may be not yet determined
• Doesn’t involve any data processing