Google Professional Machine Learning Engineer Question 296
Select 3Google Cloud PlatformYou have trained a machine learning model to predict customer churn and are now tasked with deploying it to production using Google Cloud. The model will serve thousands of predictions per second for a real-time application. Which considerations should you prioritize when serving the model using Google Cloud technologies?
- A
Ensure the serving infrastructure can scale automatically to handle variable traffic loads.
- B
Optimize the model for batch predictions to minimize latency during real-time inference.
- C
Monitor the model's performance in production for potential drift or prediction anomalies.
- D
Choose a serving platform that provides GPU support to improve inference performance for complex models.
- E
Enable manual scaling for precise control over compute resources.
Show answer and explanation
Correct answers: A, C, D
Explanation
When serving machine learning models in a real-time production environment, considerations such as autoscaling, performance monitoring, and leveraging appropriate hardware accelerators (like GPUs) are critical to ensure low-latency inference, high availability, and adaptability. Batch processing and manual scaling, on the other hand, are not suitable for real-time use cases as they do not meet the requirements for low latency and dynamic traffic management.
- A. Correct.
Correct. Autoscaling is critical for handling variable traffic loads in real-time applications to ensure consistent performance and cost-efficiency.
- B. Incorrect.
Incorrect. Batch predictions are not suitable for real-time inference as they prioritize throughput over latency, which is crucial for this use case.
- C. Correct.
Correct. Monitoring model performance in production is essential to detect issues like data drift or prediction anomalies that could degrade the model's effectiveness.
- D. Correct.
Correct. Using GPU support can improve inference performance for complex models, especially when serving thousands of predictions per second.
- E. Incorrect.
Incorrect. Manual scaling is not ideal for high-traffic, real-time applications due to its lack of flexibility and responsiveness to changing traffic patterns.