New Features
New Aggregations
Tecton 0.7 introduces support for two new built-in aggregations: percentile and count distinct, which help customers define more flexible and performant feature transformations.
Complex Data Types
This release introduces support for new complex data types: Maps, Structs, and multi-dimensional (a.k.a. nested) Arrays. These help customers create more ergonomic and performant feature definitions.
Enhanced Python Environments for On-Demand Feature Views (Public Preview)
On-Demand transformations can now be configured to run in specified Python environments which enable customers to leverage common Data Science packages when defining feature logic.
Stream Ingest API (Public Preview)
The new Tecton Stream Ingest API provides a simple endpoint to ingest existing real-time feature data, raw data streams, or data from internal services into the Tecton Feature Platform.
Changes, enhancements and resolved issues
Support for Databricks Unity Catalog
The new UnityConfig option enables customers to connect to data sources
managed by Unity Catalog, Databricks’ new unified data governance solution.
This gives Databricks customers a centralized interface for data assets,
fine-grained access control, data lineage, improved data sharing, and other new
capabilities. See
the documentation
for instructions.
Improvements to Manually-Triggered Materialization
Tecton 0.7 introduces the new manual_trigger_backfill_end_time parameter on
Feature Views configured for manual materialization — this will automatically
backfill the Feature View until the specified timestamp. See
the documentation
for more details.
New CLI Capabilities
Tecton’s CLI now supports autocompleting commands (by pressing tab). Run
tecton completion -h to get started and see
the documentation
for more details.
Users can now also use the CLI to invite users and manage ACL roles in bulk. Run
tecton user invite -h or tecton access-control assign-role -h for
instructions.
High-Uptime Stream Updates
This release significantly improves uptime for streaming clusters during routine maintenance and updates, helping ensure that streaming features stay up-to-date.
Unit Testing Improvements
Tecton’s unit testing framework now supports testing feature retrieval. Users
can pass mock data into get_historical_features() with the mock_inputs
parameter and then test the resulting outputs.
The run() method now also supports unit testing via the mock_inputs
parameter. See the
documentation
for more details.
Minor Feature View parameter changes
Tecton allows customers to specify a ttl in Feature View definitions, which
determines 1) how long features will live in the online store and 2) how
far to “look back” relative a training event’s timestamp when generating offline
training data. In Tecton 0.7, ttl is set to None by default. When a Feature
View has ttl unspecified or set to None, 1) feature data will not expire
from the online store and 2) the “look back” limit for offline training data
generation will be the feature start time.
In addition, some Feature View tags and parameters were added, renamed, deprecated, or removed as part of 0.7. Please review the Upgrade Guide for more details.
Upgrading to 0.7
See the Upgrade Guide for instructions and details on all breaking and non-breaking changes.
