ML platform
SageMaker Model Monitor
Monitors deployed models for drift and quality issues.
Key points
- Continuously monitors deployed SageMaker models for data quality and model quality issues.
- Can compare live inference data with a baseline created from training or validation data.
- Helps detect drift, missing values, schema changes, and unexpected input distributions.
- Can emit metrics and alerts so teams know when retraining or investigation is needed.
- Focuses on deployed model behavior after production release.
When to use it
- Choose Model Monitor when a deployed model may degrade over time.
- Use it to detect data drift after customer behavior, seasonality, or source systems change.
- Use it when production monitoring and alerts are the exam clues.
Exam tips
- Model Monitor detects drift; Clarify explains predictions and bias.
- Monitoring does not automatically retrain every model unless a workflow is built around it.
- Use CloudWatch-style alerting concepts for operational visibility, but the service clue is SageMaker Model Monitor.
- Ground Truth labels data before or during workflows; Model Monitor watches deployed models.