Databricks Machine Learning Professional Question 136
Select 3A data engineering team has precomputed batch predictions using a machine learning model for a recommendation system. The data science team is evaluating whether to directly query these precomputed predictions for live serving instead of deploying the model for real-time inference. Which of the following are benefits of querying precomputed batch predictions for live serving?
- A
Reduced latency during live serving of predictions
- B
Elimination of the need to retrain the model
- C
Ability to handle higher traffic loads with fewer resources
- D
Improved accuracy of predictions during live serving
- E
Simplified infrastructure requirements for production deployment
Show answer and explanation
Correct answers: A, C, E
Explanation
Querying precomputed batch predictions for live serving provides significant operational advantages, such as reducing latency, lowering resource usage to handle higher traffic, and simplifying production infrastructure. However, it does not influence the model's accuracy or its retraining needs, as these are determined by the model's design and monitoring processes.
- A. Correct.
Reduced latency is a key benefit of using precomputed batch predictions, as querying stored predictions is faster than running live model inference.
- B. Incorrect.
Retraining the model is unrelated to whether predictions are served live or precomputed; this is determined by model monitoring and performance over time.
- C. Correct.
Querying precomputed predictions reduces computational load, allowing the system to handle higher traffic with fewer resources compared to live inference.
- D. Incorrect.
The accuracy of predictions depends on the model and data, not the serving method. Precomputed predictions do not inherently improve accuracy.
- E. Correct.
Using precomputed predictions can simplify infrastructure by removing the need for low-latency model hosting and real-time processing pipelines.