Databricks Machine Learning Professional Question 126
Select 3You are tasked with deploying a trained machine learning model on Databricks to predict customer churn risk scores for a batch of customer data stored in a Delta table. The predictions need to be written back to another Delta table for downstream analysis. Which of the following steps should be included in your deployment pipeline to ensure a proper batch inference workflow?
- A
Load the input data from the Delta table into a Spark DataFrame.
- B
Use a real-time streaming job to process the input data and generate predictions.
- C
Apply the trained model to the DataFrame to generate predictions.
- D
Save the predictions back to a Delta table for further usage.
- E
Deploy the model as a REST API endpoint for batch processing.
Show answer and explanation
Correct answers: A, C, D
Explanation
A proper batch inference workflow in Databricks involves reading input data from a Delta table, applying the trained model to generate predictions, and saving the results back to another Delta table. This approach aligns with the requirements of batch processing, which deals with static datasets and prioritizes scalability and integration with storage systems like Delta Lake. Real-time approaches such as using streaming jobs or REST API endpoints are not suitable for batch-based deployments.
- A. Correct.
Correct: In a batch deployment, the input data is typically read from a storage layer (e.g., a Delta table) into a Spark DataFrame for processing.
- B. Incorrect.
Incorrect: A real-time streaming job is not suitable for batch inference workflows, as batch processing deals with static datasets rather than continuous streams of data.
- C. Correct.
Correct: Applying the trained model to the input DataFrame is essential to perform batch predictions.
- D. Correct.
Correct: Writing the predictions to a Delta table ensures the results are stored and accessible for downstream analysis or reporting.
- E. Incorrect.
Incorrect: Deploying the model as a REST API endpoint is a pattern used for real-time inference, not for batch processing.