GCP Professional Cloud Architect Practice Question

A fintech company is rebuilding its on-premises ML workflow as a Vertex AI Pipeline. The solution must 1) ingest raw transaction logs from Cloud Storage and perform cleansing and feature engineering, 2) train an XGBoost model on the processed data, 3) calculate AUC on a hold-out set and skip deployment when the metric is below 0.95, and 4) push the model to a Vertex AI endpoint for low-latency predictions when the threshold is met. Which ordered sequence of pipeline components best satisfies these requirements?

Hyperparameter-tuning component → Deployment component → Training component → Evaluation component
Training component → Data-preprocessing component → Deployment component → Evaluation component
Data-preprocessing component → Training component → Evaluation component gated by a condition → Deployment component
Data-preprocessing component → Deployment component → Evaluation component → Training component

Report Issue

Answer Description

The pipeline should first transform the raw data so that downstream steps receive a clean, well-defined dataset. Next, the training component consumes that dataset and outputs a model artifact. An evaluation component then computes the AUC and exposes the metric as an output parameter. A Condition step can read this parameter; only when it is ≥ 0.95 does it invoke the deployment component that uploads the model and creates or updates an endpoint. This pattern (preprocess → train → evaluate → conditional deploy) follows Google's recommended pipeline design and cleanly separates concerns, making each component reusable. The other sequences either deploy an unevaluated model, run steps in a non-functional order, or omit required evaluation before deployment.

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.