Google Professional Machine Learning Engineer Question 295
Single answerGoogle Cloud PlatformYou are working as a Machine Learning Engineer at a retail company. You have trained a recommendation system model and are preparing to deploy it. The model needs to serve real-time predictions during customer interactions, and the system must handle unpredictable traffic spikes during major sales events. Which deployment approach on Google Cloud would best meet these requirements?
- A
Deploy the model on AI Platform Prediction with manual scaling.
- B
Deploy the model on AI Platform Prediction with auto-scaling.
- C
Use a preemptible VM instance to host the model and manage traffic manually.
- D
Deploy the model on Cloud Run with a containerized serving solution.
Show answer and explanation
Correct answer: B
Explanation
AI Platform Prediction with auto-scaling is the best fit for serving real-time predictions during customer interactions, especially when traffic is unpredictable. Auto-scaling ensures that the system can dynamically allocate resources based on the demand, while AI Platform Prediction provides ML-specific capabilities, such as model versioning and monitoring, which are critical for production-grade ML systems.
- A. Incorrect.
Manual scaling is not ideal for unpredictable traffic because it requires pre-configured settings, which might not handle traffic spikes effectively.
- B. Correct.
Auto-scaling on AI Platform Prediction dynamically adjusts resources based on traffic, making it suitable for handling unpredictable spikes during events like sales.
- C. Incorrect.
Preemptible VM instances are cost-effective but are not reliable for real-time serving as they can be terminated without warning, leading to service interruptions.
- D. Incorrect.
Cloud Run can manage containerized workloads and auto-scale, but it is less optimized for machine learning model serving compared to AI Platform Prediction, which provides additional ML-specific features like versioning and monitoring.