Scalable data science

• All features in a few (or just one) wide tables for features and at the same time calculated in multiple notebooks
• Data scientists doing data science and not engineering (orchestration, optimized writes)
• Data scientists using SQL or PySpark
• The same feature code is used for training and inference
• Features easily reusable across many use cases by multiple teams
• Always up to date business and technical feature documentation
• Production ready code independent of particular developers

Solution?

• Feature store + Feature engineering framework

What is a feature store in general and what are the benefits?

• This article sums it up pretty well

What are the benefits of Databricks Feature store?

• Reasons listed in documentation for DBX FS

Differences/synergy of Databricks Feature store and ODAP

FAQ

Main benefits of using ODAP Feature factory

Feature factory overview