Certified Machine Learning ProfessionalDemo

By Databricks · Question Mode

QuestionQ187

Model Lifecycle Management

A Machine Learning Engineer must establish CI/CD for a new model from the Data Science team. Because the model is small and trains quickly, they use the recommended deploy-code method.

They must determine what happens when a pull request is created against the main branch from the dev workspace, which triggers the CI/CD workflow. Both the staging and production workspaces have data available for test execution. They want to use the recommended staging-workspace process within this CI/CD workflow.

Which approach accomplishes this?

  • A CI triggers unit tests and integration tests in the staging workspace, and the model is copied from the dev workspace to the staging workspace; only model deployment, inference and monitoring are integration tested.
  • B CI triggers unit tests in staging workspace and integration tests in production workspace.
  • C CI triggers unit tests and integration tests in the staging workspace, including model training, validation, deployment, inference and monitoring.
  • D CI triggers unit tests and integration tests in the staging workspace, including model training, validation; with deployment, inference and monitoring left for production workspace.
Explanation

In the recommended deploy-code workflow, staging is the CI test environment. CI runs unit tests and integration tests there, and the integration tests exercise the full ML pipeline: training, validation, deployment, inference, and monitoring. Training the model in staging is appropriate for a small, fast-training model, while production subsequently runs the approved code against production data.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!