Google Professional Machine Learning Engineer Question 421
Select 3Google Cloud PlatformYou are a Professional Machine Learning Engineer tasked with deploying a machine learning model to Google Cloud using a CI/CD pipeline. The pipeline must ensure that only models meeting performance thresholds in a staging environment are deployed to production. Which combination of steps should you include in your CI/CD pipeline to achieve this?
- A
Use Cloud Build to automate the pipeline, including training and testing the model.
- B
Integrate Vertex AI Model Monitoring to track model performance post-deployment.
- C
Implement a validation step in the pipeline to test the model on a holdout dataset in a staging environment.
- D
Deploy the model directly to production without validation to save time.
- E
Set up a manual approval step in the pipeline for human verification before production deployment.
Show answer and explanation
Correct answers: A, C, E
Explanation
To deploy a machine learning model using a CI/CD pipeline on Google Cloud, it is essential to automate the workflow with tools like Cloud Build, validate the model in a staging environment before production, and optionally include a manual approval step for additional verification. These steps ensure the model is both performant and reliable upon deployment while adhering to best practices for CI/CD in ML workflows.
- A. Correct.
Cloud Build is a managed CI/CD service in Google Cloud that allows you to automate workflows, including model training and testing, making it essential for automating the pipeline.
- B. Incorrect.
Vertex AI Model Monitoring is useful after deployment to monitor the model in production, but it is not directly related to the CI/CD pipeline's deployment and validation steps.
- C. Correct.
Validating the model in a staging environment using a holdout dataset ensures the model meets performance thresholds before production deployment, which is a critical step.
- D. Incorrect.
Skipping validation and deploying directly to production is risky and does not align with best practices for CI/CD pipelines in ML workflows.
- E. Correct.
A manual approval step provides an additional layer of verification, ensuring that only models reviewed by a human are deployed to production, which is a best practice in sensitive use cases.