Data Lineage Across the Feature Store and Driverless AI
- Observability
- H2O Driverless AI
- H2O Feature Store
This lesson explains how data lineage works across the Feature Store and Driverless AI. In the Feature Store, teams collaborate on version controlled feature sets, and the platform tracks which feature set or version a later one, including a derived one, came from. In Driverless AI, once a model is built, its automatic documentation captures exactly which data was used, its distributions, and where it lives on the system. The lesson closes by noting that this lineage carries through to deployment, so a model version can always be traced back to the experiment and training dataset it came from.
Key takeaways
- How version control in the Feature Store tracks which feature set or version a later one was derived from
- What Driverless AI's automatic model documentation captures for each experiment: the exact data used and its distributions
- How a deployed model can be traced back to its model version, the experiment that built it, and its training dataset
- Why knowing exactly where a dataset lives on the system supports full data lineage regardless of where it is used
Start owning your AI