Databricks Machine Learning Professional Question 139
Select 3A data science team is building a machine learning solution for real-time fraud detection on a financial transaction dataset. To handle high query volumes with low latency, they decide to serve precomputed batch predictions using a feature store. What are the benefits of querying precomputed batch predictions in this scenario?
- A
Reduced serving latency since predictions are already precomputed
- B
Increased prediction accuracy due to precomputed batch processing
- C
Lower computational cost during live serving
- D
Simplified infrastructure for handling large-scale requests
- E
Ability to dynamically retrain the model during live serving
Show answer and explanation
Correct answers: A, C, D
Explanation
Querying precomputed batch predictions is beneficial in scenarios requiring low latency and high scalability, such as real-time fraud detection. By precomputing predictions during a batch process, the system reduces serving latency, lowers computational cost at query time, and simplifies the infrastructure needed to handle large-scale requests. However, this approach does not improve model accuracy or allow for dynamic retraining during live serving.
- A. Correct.
Precomputed batch predictions allow the system to retrieve results without running the model in real-time, significantly reducing serving latency.
- B. Incorrect.
Prediction accuracy is not inherently improved by precomputing predictions; it depends on the quality of the model and data, not the serving method.
- C. Correct.
Since predictions are precomputed, the system does not need to utilize compute resources for real-time model inference, reducing the computational cost.
- D. Correct.
Precomputing predictions simplifies the serving infrastructure by offloading model inference to the batch processing phase, making it easier to scale for large request volumes.
- E. Incorrect.
Precomputing predictions does not enable dynamic retraining. Model retraining is a separate process and unrelated to how predictions are served.