Customized model monitoring for near real-time batch inference with Amazon SageMaker

This article discusses customized model monitoring for near real-time batch inference with Amazon SageMaker. It presents a framework to handle multi-payload inference requests for near real-time inference scenarios.

Specifically, the article covers:

• Overview of the solution architecture
• Prerequisites for following along
• Steps to train an XGBoost model
• Creating custom inference code to handle multi-payload requests
• Deploying a SageMaker endpoint with data capture enabled
• Creating constraints for model quality monitoring
• Publishing a custom Docker image for model monitoring
• Creating a SageMaker Model Monitor schedule with the custom image
• Observing the model monitoring job output and violations
• Cleaning up resources