Question

You deployed a credit-scoring model to an Azure Machine Learning managed online endpoint. The compliance team requires that you capture the model's input requests and output predictions to a datastore so they can later analyze production behavior and feed a monitoring pipeline. You want to enable this with minimal custom code. What should you configure on the deployment?

Accepted Answer

Enable data collection on the online deployment so request and response payloads are logged to blob storage.

Answer

Add Application Insights logging statements in the scoring script to print each prediction to the trace log.

Answer

Increase the instance count on the deployment so more telemetry is emitted to the endpoint metrics.

Answer

Register the endpoint's traffic as a versioned data asset after each invocation.