Enable different model versions and dataset versions to have different dependent variables / target features.
Added support for grid based models or lookup models in Model Registry. These can be imported from CSV and directly modified on the platform.
Allow Models to be registered without a dependent variable. This makes it easier to create models which are simple formulas and have no dependent variable. In such models, the Model Quality reports are not generated by default.
Spark support- Added ability to write aggregate definitions in spark across the platform.
Removed the deprecated def score method of writing model definitions for python-function models. (Need to document in backward compatibility)
Common
Reviews- Added the ability to do periodic ongoing reviews on an approved object to enable cases where models need to be reviewed at intervals to ensure they continue to provide valid results.
Ongoing reviews can be done manually similar to the existing approval process
Added concept of scheduled review, where a user can schedule a review in the future.
Platform Entities-Enable clients to customize the set of Platform Entities that need to be used in the Platform. Currently only unused Entities can be edited, and new Entities can be created in the Settings module.
Role Management- Added the “superuser” concept in Role Management, which is an advanced functionality to enable certain users to view all objects - including Drafts from other users.
Grouping- Add groups to all registry pages consistently. Added groups for: Table Registry, Quality Profile, Policy Registry, Global Variables, Runtime Parameters, Global Functions, Product Registry, Product Types and Framework Registry.
Technical
Added a CLI script to share objects to users programmatically.