Databricks Machine Learning Professional Question 19
Select 3You are working on a machine learning pipeline using the Databricks Feature Store. You need to track and update feature data for a model training workflow. Specifically, you must append new feature data to the existing Feature Store table without overwriting existing records. After updating the table, you also want to ensure the updated feature data is linked to a new experiment run for tracking purposes. Which combination of actions should you take?
- A
Use the 'write_table' method with the 'overwrite' mode to add new feature data to the Feature Store table.
- B
Use the 'merge' method in the Feature Store to append new feature data to the existing table.
- C
Log the experiment run using MLflow and associate it with the updated feature data version.
- D
Use the 'read_table' method to verify the updated Feature Store table after merging the data.
- E
Delete the previous versions of the Feature Store table to avoid conflicts with the experiment run.
Show answer and explanation
Correct answers: B, C, D
Explanation
The correct combination of actions involves using the 'merge' method to append new feature data without overwriting existing records, logging the experiment run in MLflow to track and associate the updated feature data, and verifying the updated table using the 'read_table' method. These steps ensure that the Feature Store table is correctly updated, the changes are tracked, and the data integrity is maintained.
- A. Incorrect.
The 'write_table' method with 'overwrite' mode would replace the existing data in the Feature Store table, which contradicts the requirement to append new feature data without overwriting.
- B. Correct.
The 'merge' method is the correct approach for appending new data to an existing Feature Store table while preserving the existing records.
- C. Correct.
Logging the experiment run using MLflow ensures that the updated feature data is tracked and associated with the new experiment, which is essential for reproducibility and tracking in machine learning workflows.
- D. Correct.
Using the 'read_table' method allows you to confirm that the new feature data has been successfully merged into the existing Feature Store table, which is a recommended practice.
- E. Incorrect.
Deleting previous versions of the Feature Store table is not necessary and could lead to loss of data lineage and reproducibility, which are critical in machine learning workflows.