GCP Professional Cloud Architect Practice Question

A retailer runs a nightly Vertex AI Pipeline that extracts sales data from BigQuery, processes it in Dataflow, trains an XGBoost model on pre-emptible A100 GPUs, registers the model, and can deploy it to production. The team must (1) capture complete lineage and evaluation metrics for every run and (2) automatically deploy only when the new model's RMSE is at least 3 % lower than the current production model-without building extra services. Which Vertex AI capability best meets these requirements?

Using Vertex AI Experiments to tag each model version and manually review RMSE before approving deployment
Exporting training logs to Cloud Logging and invoking a Cloud Function that parses the logs and decides whether to deploy the new model
Leveraging Vertex AI Pipelines' automatic ML metadata tracking with a conditional execution step that deploys only if the new run's recorded rmse metric improves by at least 3 %
Configuring Vertex AI continuous evaluation to trigger alerts when model performance changes

Report Issue

Answer Description

Vertex AI Pipelines automatically writes execution details-datasets, code versions, parameters, and metrics-into Vertex AI's ML metadata store. The pipeline definition can include a conditional step that compares a metric (for example, the rmse output of the evaluation component) with the metric stored for the previously deployed model and proceeds to the deployment step only if the required 3 % improvement is met. This provides end-to-end lineage, reproducibility, and policy-based promotion inside the managed pipeline service, eliminating the need for separate logging systems or external functions.

Using Vertex AI Experiments alone records metrics but still relies on manual review; it does not automate conditional deployment. Continuous evaluation focuses on serving-time data drift rather than training-time metric comparison. Building custom Cloud Logging parsers and Cloud Functions would satisfy the requirement but contradicts the constraint to avoid extra services and lacks built-in lineage management.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.