Your team is building an automated continuous training (CT) and continuous deployment (CD) pipeline for a mission-critical tabular classification model on Google Cloud. You need to implement a solution using Vertex AI Pipelines that continuously evaluates candidate models and deploys them to production. The pipeline implementation must meet the following operational requirements: Track end-to-end dataset lineage, hyperparameter configurations, and model artifact provenance in Vertex ML Metadata. Execute automated model evaluation on validation test data and compute classification metrics. Use conditional execution logic within the pipeline directed acyclic graph (DAG) to trigger deployment to the production endpoint only if candidate model metrics exceed a target accuracy threshold. Which architecture and pipeline design should you implement?

Author a Kubeflow pipeline with the Google Cloud Pipeline Components SDK using `ModelEvaluationClassificationOp` to generate evaluation metrics, log metadata to Vertex ML Metadata, and gate deployment to a `google.VertexEndpoint` via `ModelDeployOp` inside a `dsl.Condition` block.: ### What the Solution Is This architecture uses **Vertex AI Pipelines** orchestrated via the **Kubeflow Pipelines (KFP) SDK** and **Google Cloud Pipeline Components (GCPC)**. It integrates automated model evaluation using `ModelEvaluationClassificationOp`, logs all pipeline artifacts natively into **Vertex ML Metadata**, and enforces a quality deployment gate using native KFP conditional logic (`dsl.Condition`) before invoking `ModelDeployOp`. ### How It Meets the Requirements - **Artifact Lineage and Metadata Telemetry**: Running within **Vertex AI Pipelines** automatically registers all inputs, intermediate datasets, execution parameters, and model outputs into **Vertex ML Metadata**. First-party artifact types such as `google.VertexModel` and `google.VertexEndpoint` preserve end-to-end data and model lineage. - **Automated Model Evaluation**: `ModelEvaluationClassificationOp` ingests ground-truth test data and candidate batch prediction outputs to compute standard classification metrics (such as ROC-AUC, accuracy, precision-recall, and sliced metrics). - **Conditional Deployment Gating**: Placing the `ModelDeployOp` component inside a `dsl.Condition` block allows the pipeline DAG to dynamically evaluate whether the metric output meets or exceeds the required threshold. The model is uploaded via `ModelUploadOp` and deployed to the existing target `google.VertexEndpoint` only if the condition evaluates to true. ### Key Advantages and Capabilities - **Serverless Automation**: **Vertex AI Pipelines** manages execution infrastructure without requiring persistent compute instances. - **Native Lineage Tracking**: Provides full visibility into model provenance, training data snapshots, and performance comparison across historical pipeline runs. - **Deterministic Promotion**: Prevents degraded models from reaching live traffic by enforcing machine-checked gates within the pipeline definition. - **First-Party Interoperability**: Seamlessly passes typed artifacts (`google.VertexModel`) across components without custom boilerplate. ### Why This Is Optimal This approach aligns with Google Cloud recommended MLOps practices. It eliminates manual intervention, leverages managed components designed specifically for tabular classification evaluation, and avoids brittle external glue code or post-deployment rollback scripts.

Vertex Ai Continuous Training Evaluation Conditional Deployment | Professional Cloud DevOps Engineer

Vertex Ai Continuous Training Evaluation Conditional Deployment

Worked example. The correct answer is already marked and every option is explained below, so there is nothing to select here. To answer questions yourself, start the free trial.

Your team is building an automated continuous training (CT) and continuous deployment (CD) pipeline for a mission-critical tabular classification model on Google Cloud. You need to implement a solution using Vertex AI Pipelines that continuously evaluates candidate models and deploys them to production.

The pipeline implementation must meet the following operational requirements:

Track end-to-end dataset lineage, hyperparameter configurations, and model artifact provenance in Vertex ML Metadata.
Execute automated model evaluation on validation test data and compute classification metrics.
Use conditional execution logic within the pipeline directed acyclic graph (DAG) to trigger deployment to the production endpoint only if candidate model metrics exceed a target accuracy threshold.

Which architecture and pipeline design should you implement?

Author a Kubeflow pipeline with the Google Cloud Pipeline Components SDK using ModelEvaluationClassificationOp to generate evaluation metrics, log metadata to Vertex ML Metadata, and gate deployment to a google.VertexEndpoint via ModelDeployOp inside a dsl.Condition block.

Deploy the candidate model immediately to the production google.VertexEndpoint using ModelDeployOp, and configure a post-deployment ModelEvaluationForecastingOp step to automatically initiate an endpoint rollback if live inference accuracy drops below the threshold.

Stream pipeline training metrics into Cloud Logging, configure a log-based metric to trigger an external Cloud Run function via Pub/Sub, and have the function parse logs to invoke ModelUploadOp via the REST API while bypassing Vertex ML Metadata.

Stream candidate model predictions to an AlloyDB database using an Apache Beam pipeline, calculate classification metrics inside AlloyDB SQL stored procedures, and invoke VertexAITextEmbeddings to update endpoint configurations.

Professional Cloud DevOps Engineer

Vertex Ai Continuous Training Evaluation Conditional Deployment

Related questions