Databricks Machine Learning Professional Question 130
Single answerA data engineering team has developed a machine learning model to predict customer churn for a subscription-based service. The predictions will be used to identify customers who are likely to cancel their subscriptions in the next 30 days. The team needs to deploy the model to generate predictions for all customers once a week. Which deployment approach is the most appropriate for this use case?
- A
Online deployment with low-latency REST API endpoints
- B
Batch deployment that processes all customer data weekly
- C
Real-time streaming deployment for continuous predictions
- D
Edge deployment to run the model directly on customer devices
Show answer and explanation
Correct answer: B
Explanation
Batch deployment is the most suitable approach for the described use case because the predictions are generated for the entire customer base on a fixed schedule (once a week). Batch deployment efficiently handles such periodic, large-scale inference tasks, whereas online, streaming, or edge deployments would be unnecessary and potentially more complex to implement for this scenario.
- A. Incorrect.
Online deployment with low-latency REST API endpoints is suitable for real-time, on-demand predictions. However, in this case, predictions are needed for the entire dataset on a weekly schedule, not in real time.
- B. Correct.
Batch deployment that processes all customer data weekly is the correct choice because the requirement is to generate predictions for the entire customer base on a fixed weekly schedule, which aligns well with the batch processing paradigm.
- C. Incorrect.
Real-time streaming deployment is used for continuously incoming data streams, where predictions are required instantly. This is unnecessary in the given use case since the predictions are generated on a weekly basis.
- D. Incorrect.
Edge deployment is designed for scenarios where models need to run locally on devices, typically in environments with limited connectivity. This is irrelevant to the described use case as the predictions are generated centrally for all customers.