Databricks Machine Learning Professional Question 138
Select 3A retail company uses a machine learning model to predict customer churn. To improve live serving latency, they decide to use precomputed batch predictions instead of real-time model inference. What are the key benefits of querying precomputed batch predictions in this scenario?
- A
Reduced serving latency during live queries
- B
Elimination of the need for model retraining
- C
Scalability improvements for handling large query volumes
- D
Ability to provide model predictions for historical data analysis
- E
Reduced computational cost during live serving
Show answer and explanation
Correct answers: A, C, E
Explanation
Precomputed batch predictions significantly enhance live serving by reducing latency and computational demand during queries. This approach shifts the intensive computation to the batch processing phase, making live systems more scalable and cost-efficient. While this method does not eliminate the need for retraining or directly impact historical data analysis, it is particularly effective for live workloads that prioritize speed and efficiency.
- A. Correct.
Reduced serving latency during live queries is a benefit because precomputed predictions are stored and readily available, avoiding the need for real-time model computation.
- B. Incorrect.
Elimination of the need for model retraining is incorrect because precomputing predictions does not impact the retraining process, which depends on data and model maintenance cycles.
- C. Correct.
Scalability improvements for handling large query volumes is correct because querying a database or storage system with precomputed predictions is generally more scalable than performing real-time inference for each query.
- D. Incorrect.
Ability to provide model predictions for historical data analysis is incorrect as historical data analysis is not directly related to the live serving benefits of precomputed batch predictions.
- E. Correct.
Reduced computational cost during live serving is correct because precomputing predictions shifts the computational load to the batch processing phase, reducing costs during live serving.