Question

Your team trained an MLflow model in Azure Machine Learning that consumes features materialized in a managed feature store. Data scientists retrieved the training features using a feature retrieval specification generated from the feature store. To ensure the same feature transformations and lookups are applied automatically at inference time when the model is deployed to an online endpoint, what should you do before registering the model?

Accepted Answer

Package the feature retrieval specification into the model artifact folder so it is registered together with the MLflow model.

Answer

Store the feature retrieval specification as a separate data asset and reference it manually in the scoring script at deployment time.

Answer

Recreate the feature transformations inline within the endpoint's scoring script so no feature store dependency is required.

Answer

Register the feature set version as a model dependency in the environment's conda file.