Databricks Machine Learning Associate Question 445
Select 3A data science team has trained a machine learning model and saved it as a registered model in the Databricks Model Registry. They now want to score the model using features stored in a Delta table managed by Databricks Feature Store. Which of the following steps are required to score the model correctly?
- A
Use the FeatureStoreClient to retrieve the features from the feature store table.
- B
Load the model as a PyFunc model from the Databricks Model Registry.
- C
Directly access the Delta table using Spark SQL without involving the FeatureStoreClient.
- D
Ensure the feature table contains a primary key matching the input data for scoring.
- E
Use the FeatureStoreClient to write the scoring results back to the feature store table.
Show answer and explanation
Correct answers: A, B, D
Explanation
Scoring a model using features from a Databricks Feature Store table involves retrieving the features using the FeatureStoreClient, ensuring the feature table has a matching primary key for input data, and loading the model from the registry as a PyFunc model to apply it for scoring. Directly accessing the Delta table or writing results back to the feature store table are not part of the standard workflow.
- A. Correct.
Correct. The FeatureStoreClient is required to retrieve features from the feature store table, ensuring the feature transformations and joins are applied appropriately.
- B. Correct.
Correct. To use the registered model for scoring, it must be loaded as a PyFunc model from the Databricks Model Registry.
- C. Incorrect.
Incorrect. Directly querying the Delta table bypasses feature store-specific functionality like feature transformations and primary key validation, which are essential for correct scoring.
- D. Correct.
Correct. The feature table must have a primary key that aligns with the input data to ensure the correct features are retrieved for scoring.
- E. Incorrect.
Incorrect. The FeatureStoreClient is not used to write scoring results back to the feature store table. Typically, scoring results are stored in separate tables or files.