Automatic feature lookup with MLflow models on Databricks
Model Serving can automatically look up feature values from published online stores.
Requirements
The model must have been logged with
FeatureEngineeringClient.log_model
(for Feature Engineering in Unity Catalog) orFeatureStoreClient.log_model
(for Workspace Feature Store, requires v0.3.5 and above).The online store must be published with read-only credentials.
Note
You can publish the feature table at any time prior to model deployment, including after model training.
Automatic feature lookup
Databricks Model Serving supports automatic feature lookup from these online stores:
Amazon DynamoDB (v0.3.8 and above)
Automatic feature lookup is supported for the following data types:
IntegerType
FloatType
BooleanType
StringType
DoubleType
LongType
TimestampType
DateType
ShortType
DecimalType
ArrayType
MapType
Override feature values in online model scoring
All features required by the model (logged with FeatureEngineeringClient.log_model
or FeatureStoreClient.log_model
) are automatically looked up from online stores for model scoring. To override feature values when scoring a model using a REST API with Model Serving include the feature values as a part of the API payload.
Note
The new feature values must conform to the feature’s data type as expected by the underlying model.
Notebook examples: Unity Catalog
With Databricks Runtime 13.2 and above, any Delta table in Unity Catalog with a primary key can be used as a feature table. When you use a table registered in Unity Catalog as a feature table, all Unity Catalog capabilities are automatically available to the feature table.
This example notebook illustrates how to publish features to an online store and then serve a trained model that automatically looks up features from the online store.