Question

Your team runs a Vertex AI Pipeline that trains a fraud-detection model nightly. Business stakeholders require that a newly trained model is only deployed to the production endpoint if its evaluation AUC exceeds the currently deployed model's AUC by at least 0.02; otherwise the pipeline should skip deployment and leave the existing model serving. You want to implement this entirely within the pipeline definition using the KFP v2 SDK. Which approach best meets this requirement?

Accepted Answer

Add an evaluation component that outputs the metric, then use a dsl.Condition block that gates the deployment component on the comparison between the new and current AUC values.

Answer

Deploy every trained model to the endpoint, then rely on Vertex AI Model Monitoring to automatically roll back models whose AUC does not improve.

Answer

Schedule two separate pipelines—one for training and one for deployment—and manually inspect Vertex Experiments each morning before triggering the deployment pipeline.

Answer

Set the deployment component's caching to enabled so that it is only re-executed when the evaluation metric artifact changes.