Model Registry & Monitoring — Vertex AI MLOps

[ ABORT TO HUD ]
SEQ. 1
SEQ. 2

Model Registry & Monitoring

🔄 Vertex AI MLOps 15m 300 BASE XP⌨ HANDS-ON LAB

Enterprise Governance & Continuous Data Drift Detection

Once a machine learning or fine-tuned foundation model passes automated evaluation gates, it is registered in the Vertex AI Model Registry. The Registry serves as the central control plane for versioning (v1, v2, candidate, production), deployment aliases, and regulatory approval gates.

Following deployment to online prediction endpoints, Vertex AI Model Monitoring continuously samples live incoming traffic and compares feature distributions against training baseline datasets.

Statistical Drift Metrics
  • Population Stability Index (PSI): Flags categorical and continuous feature distribution divergence between baseline and live requests.
  • Kolmogorov-Smirnov (KS) Test: Statistical test detecting numeric distribution drift that often indicates changing customer behavior or upstream pipeline failures.
  • Automated Retraining Triggers: Emits Cloud Monitoring alerts and Eventarc events to automatically trigger a Vertex Pipeline when drift exceeds predefined thresholds.
from google.cloud import aiplatform

# Register a production model artifact into Vertex Model Registry
model = aiplatform.Model.upload(
    display_name="fraud_detection_xgb",
    artifact_uri="gs://ml-dw/models/fraud_v3/",
    serving_container_image_uri="us-docker.pkg.dev/vertex-ai/prediction/xgboost-cpu.1-6:latest",
    description="Production XGBoost fraud classifier with continuous drift monitoring",
    version_aliases=["production", "champion"]
)
print(f"Registered model resource: {model.resource_name}")
⌨ HANDS-ON LABList Vertex AI Model Monitoring Jobs
⭐ +150 XP

Audit automated model monitoring jobs configured to alert on feature drift and prediction skew.

1List active model monitoring jobs in the us-central1 region.
lab-sandbox — simulated environment
INFINITY LAB SANDBOX v2.6 — simulated shell
Type the command for the current objective. Helpers: "hint", "solution", "clear".
$
OBJECTIVE 1 / 1 — type "hint" if stuck
SYNAPSE VERIFICATION
QUERY 1 // 2
What phenomenon does Vertex AI Model Monitoring attempt to detect?
Data Drift (when live data diverges from training data)
Cybersecurity breaches
Website downtime
Hardware failures