Databricks Machine Learning Associate Question 442
Select 3You have trained a machine learning model to predict customer churn and stored relevant features in a Databricks Feature Store table. Now, you want to score the model using the features from the feature store table for a batch of new customer IDs. Which of the following steps are required to correctly score the model?
- A
Fetch the features from the feature store using the customer IDs and join them with the batch data.
- B
Register the model in the Databricks Model Registry before scoring.
- C
Use the
FeatureStoreClientto ensure feature lookups are consistent with the feature store. - D
Directly pass the customer IDs to the model without fetching features from the feature store.
- E
Ensure the model was trained using the same feature store table to avoid feature mismatches.
Show answer and explanation
Correct answers: A, C, E
Explanation
To score a model using features from a feature store, you need to fetch the required features for the given IDs using the FeatureStoreClient and ensure that the model and scoring process are consistent with the feature store table used during training. This ensures the predictions are accurate and based on the correct feature set.
- A. Correct.
Correct. To score the model, you need to fetch the features corresponding to the customer IDs from the feature store and join them with the batch data for predictions.
- B. Incorrect.
Incorrect. While registering the model in the Databricks Model Registry is good practice for production use, it is not a mandatory step for batch scoring from a feature store.
- C. Correct.
Correct. The
FeatureStoreClientensures that feature lookups are consistent and handles the interaction with the feature store for retrieving and managing features. - D. Incorrect.
Incorrect. You cannot directly pass customer IDs to the model without fetching features, as the model relies on the feature data for predictions.
- E. Correct.
Correct. Consistency between the training and scoring datasets is critical. Using the same feature store table ensures no feature mismatches occur during scoring.