Google Professional Machine Learning Engineer Question 294
Select 2Google Cloud PlatformYou are deploying a machine learning model on Google Cloud for real-time predictions. The model requires low latency, high scalability, and periodic updates. Which of the following considerations should you prioritize when serving the model?
- A
Use Vertex AI Prediction with automatic scaling to handle traffic spikes.
- B
Deploy the model on a Compute Engine instance with a pre-configured GPU for low latency.
- C
Leverage a managed service like Cloud Run to serve the model as a containerized application.
- D
Enable model versioning in Vertex AI Prediction for seamless updates.
- E
Choose BigQuery ML for serving the model due to its SQL integration.
Show answer and explanation
Correct answers: A, D
Explanation
For serving a machine learning model requiring low latency, high scalability, and periodic updates, Vertex AI Prediction is the most effective solution. It provides automatic scaling for traffic spikes and model versioning for seamless updates. Compute Engine and Cloud Run, while useful in some cases, require more manual management and are less optimized for real-time ML serving. BigQuery ML is not designed for real-time serving, making it unsuitable here.
- A. Correct.
Vertex AI Prediction with automatic scaling is well-suited for real-time predictions requiring low latency and high scalability. It also supports automatic scaling to handle traffic spikes efficiently.
- B. Incorrect.
While a Compute Engine instance with a GPU can provide low latency, it requires significant manual management for scalability and is less optimal for real-time serving compared to fully managed services like Vertex AI Prediction.
- C. Incorrect.
Cloud Run is a good option for containerized applications, but it may not provide the same level of optimization for machine learning models as Vertex AI Prediction, especially for features like automatic scaling and model versioning.
- D. Correct.
Enabling model versioning in Vertex AI Prediction simplifies the process of updating models and ensures minimal disruption during updates, aligning with the need for periodic model updates.
- E. Incorrect.
BigQuery ML is primarily designed for in-database model training and batch predictions rather than real-time serving, making it unsuitable for this scenario.