Google Professional Machine Learning Engineer Question 254
Select 2Google Cloud PlatformYou are a machine learning engineer tasked with deploying a tabular dataset classification model for predicting customer churn. The model was trained using AutoML Tables, and you need to deploy it in a Kubernetes Engine (GKE) cluster for scalable predictions. Which steps should you perform to ensure the model is deployed successfully and predictions can be served at scale?
- A
Export the trained AutoML Tables model to a TensorFlow SavedModel format and create a custom prediction service to deploy on GKE.
- B
Use the 'Deploy Model' option in the AutoML Tables interface to deploy the model directly to the GKE cluster.
- C
Export the trained AutoML Tables model to a Vertex AI endpoint and integrate the endpoint with your application.
- D
Containerize the AutoML Tables model using Cloud Build, then deploy the container to the GKE cluster.
- E
Use the Vertex AI Prediction service to serve the AutoML Tables model and route requests from GKE workloads to the prediction endpoint.
Show answer and explanation
Correct answers: C, E
Explanation
AutoML Tables models are designed to integrate with Vertex AI for deployment. By exporting the trained model to a Vertex AI endpoint, you can leverage Google Cloud's managed prediction infrastructure, which ensures scalability and reliability. Additionally, routing GKE workload requests to the Vertex AI Prediction endpoint is the recommended way to serve predictions at scale while maintaining flexibility in your application architecture.
- A. Incorrect.
Exporting the AutoML Tables model to a TensorFlow SavedModel format and creating a custom service is unnecessary as AutoML Tables models are designed to be deployed using managed services like Vertex AI. This step is overly complex and not recommended.
- B. Incorrect.
The 'Deploy Model' option in the AutoML Tables interface does not support direct deployment to GKE. It supports deployment to Vertex AI endpoints instead.
- C. Correct.
Exporting the AutoML Tables model to a Vertex AI endpoint is a valid approach for serving predictions. Vertex AI provides managed deployment and scalable prediction services.
- D. Incorrect.
Containerizing the AutoML Tables model using Cloud Build is not supported, as AutoML Tables models are not exported as containerized artifacts. Managed deployment services like Vertex AI should be used instead.
- E. Correct.
Using the Vertex AI Prediction service to serve the AutoML Tables model and routing GKE workload requests to the prediction endpoint is a valid and scalable approach. This allows you to leverage managed prediction services while integrating with GKE workloads.