Databricks Machine Learning Associate Question 90
Select 3You are tasked with scoring a machine learning model that predicts customer churn using features stored in a Databricks Feature Store table. The feature store table is named customer_features, and the model has already been registered in the Databricks Model Registry. Which of the following steps are required to correctly score the model using the feature store table?
- A
Use the Feature Store client to read the
customer_featurestable and join it with the model's required feature set. - B
Load the model from the Model Registry and use the Feature Store client to score the model against the features.
- C
Directly query the
customer_featurestable using Spark SQL and pass the resulting DataFrame to the model for scoring. - D
Ensure that the feature store table is materialized as a Delta table before scoring the model.
- E
Use the
fs.score_batchmethod from the Feature Store client to generate predictions.
Show answer and explanation
Correct answers: A, B, E
Explanation
Scoring a model using features from a feature store table in Databricks requires using the Feature Store client to fetch and join the required features, loading the model from the Model Registry, and using the fs.score_batch method to ensure proper integration and feature consistency. Directly querying the table or requiring explicit materialization of the table is unnecessary because the Feature Store client abstracts these operations.
- A. Correct.
Correct: The Feature Store client is required to fetch and join the features from the
customer_featurestable with the required feature set for scoring. - B. Correct.
Correct: The model must be loaded from the Model Registry, and the Feature Store client is used to correctly score the model with the features.
- C. Incorrect.
Incorrect: Directly querying the feature store table using Spark SQL bypasses the Feature Store client, which ensures proper versioning, transformations, and consistency of features.
- D. Incorrect.
Incorrect: While feature store tables are often stored as Delta tables, there is no requirement to explicitly materialize the table before scoring. The Feature Store client handles accessing the stored features.
- E. Correct.
Correct: The
fs.score_batchmethod in the Feature Store client is designed to score a batch of data using the specified model and feature store table.