Google Professional Machine Learning Engineer Question 297
Select 3Google Cloud PlatformYour team has developed a machine learning model to predict customer churn. The model is trained and ready for deployment on Google Cloud. The team is considering the best approach to serve the model with low latency and high scalability. Which of the following should you consider as part of the deployment and serving process?
- A
Use AI Platform Prediction for model serving, as it provides auto-scaling and built-in monitoring.
- B
Deploy the model on a Compute Engine virtual machine to retain full control over the serving environment.
- C
Use Cloud Functions for serving the model to achieve near-real-time inference.
- D
Implement model versioning to allow seamless updates and rollback capabilities.
- E
Leverage TensorFlow Serving on Google Kubernetes Engine (GKE) for custom deployment and scalability.
Show answer and explanation
Correct answers: A, D, E
Explanation
When serving ML models on Google Cloud, it is crucial to consider scalability, ease of deployment, and operational efficiency. AI Platform Prediction is a managed service that simplifies deployment with auto-scaling and monitoring, making it suitable for production workloads. Model versioning ensures operational flexibility, enabling updates and rollbacks without disrupting the service. TensorFlow Serving on GKE is another effective option for teams requiring custom deployment configurations with scalability. While Compute Engine and Cloud Functions may work in specific scenarios, they are generally less suitable for production-grade model serving due to their limitations in scalability, latency, and operational complexity.
- A. Correct.
AI Platform Prediction is a recommended approach for serving ML models at scale on Google Cloud. It provides features like auto-scaling, monitoring, and easy integration with other GCP services.
- B. Incorrect.
While deploying the model on Compute Engine gives full control over the serving environment, it requires significant manual effort to manage scalability, monitoring, and updates, making it less suitable for production use cases requiring low latency and high scalability.
- C. Incorrect.
Cloud Functions is not ideal for serving ML models due to limitations such as execution timeout and lack of support for GPU acceleration, which may be required for low-latency inference.
- D. Correct.
Model versioning is a critical consideration when serving models, as it allows teams to manage updates, rollbacks, and multiple versions of a model efficiently.
- E. Correct.
TensorFlow Serving on GKE is a strong option for custom deployments, offering flexibility, containerization, and scalability for serving ML models in production environments.