Databricks Machine Learning Associate Question 92
Select 3You are tasked with scoring a machine learning model using features stored in a Databricks Feature Store table. The feature table, 'customer_features', contains precomputed features for customer churn prediction. Which of the following steps are required to correctly score the model using this feature table?
- A
Register the model to the Databricks Model Registry before scoring.
- B
Use the FeatureStoreClient to retrieve the features from the feature table.
- C
Join the feature table with the input data on the required primary keys.
- D
Export the feature table to a CSV file and use it for scoring.
- E
Ensure the feature table's schema matches the model's input schema.
Show answer and explanation
Correct answers: B, C, E
Explanation
To score a model using features from a Databricks Feature Store table, you must programmatically retrieve features using the FeatureStoreClient, join the features with the input data using primary keys, and ensure the feature table's schema matches the model's input schema. These steps ensure the model receives the correct data format and feature values for accurate predictions.
- A. Incorrect.
Registering the model in the Model Registry is not strictly required for scoring a model using features from a feature store. However, it is useful for model management and deployment.
- B. Correct.
The FeatureStoreClient is a key component in Databricks for interacting with feature tables. It is required to programmatically retrieve features from the feature table.
- C. Correct.
Joining the feature table with the input data on primary keys ensures the correct features are matched to the input data. This is a crucial step in scoring.
- D. Incorrect.
Exporting the feature table to a CSV file is not a recommended or required step in the Databricks Feature Store workflow. The feature store is designed to be used programmatically within Databricks.
- E. Correct.
Ensuring the feature table's schema matches the model's input schema is essential for successful scoring, as mismatches between the feature table and model input can lead to errors.