Main features & enhancements
- Feature Store: Support controlling whether to update stats, @katyakats
- Frameworks: Add XGBoost, LGBM and SciKitLearn to MLRun's frameworks infrastructure, @guy1992l
- Frameworks: Add AutoMLRun
apply_mlrun
and replaced bokeh with plotly, @guy1992l - Frameworks: Support LGBM auto-logging, @AlxZed
- Datastore: Enable no credentials for working with GCP workload identity, @theSaarco
- Model Monitoring: Use feature store for model monitoring graph, @dinal
- Runtimes: Limit function's service accounts based on project configuration, @theSaarco
- Runtimes: Enhance Nuclio http/stream triggers + support ignore tagged cells, @yaronha
- Runtimes: When editing a function, you can force a rebuild of the image by checking the Build a new image option. (The default is not checked.), @yaronha
- Secrets: Delete project secrets on project deletion, @theSaarco
- API: Decouple the migrations from service initialization, @Hedingber
- Requirements: Bump scikit-learn to 1.x, @Hedingber
- Requirements: Support PyArrow 5, @gtopper
- Requirements: Bump pip to 21.2.x and python to 3.7.11, @Hedingber
- System Tests: Add test for remote spark ingestion, @urihoenig
- UI: When editing a function, you can force a rebuild of the image by checking the Build a new image option (The default is not checked), @ilan7empest
- UI: Release notes
Notes
- Limitations: Ingesting data: Do not name columns starting with either
t_
oraggr_
. They are reserved for internal use, and the data does not ingest correctly - Limitations: When creating a feature-vector, feature-sets cannot be referenced using a version or tag. In practice this means that only features from the latest version of a feature-set can be used. ((In the UI be careful to only select features from the latest version)