Question

A machine learning team at a healthcare analytics company runs a SageMaker Pipeline that includes processing, training, and evaluation steps. Auditors require that for every deployed model, the team can trace back to the exact dataset version, processing code, and training hyperparameters used to produce it. The team wants this lineage captured automatically without building custom logging infrastructure. Which approach best meets this requirement?

Accepted Answer

Rely on SageMaker ML Lineage Tracking, which automatically records the entities and relationships (data, code, and steps) created and consumed by SageMaker Pipeline executions.

Answer

Add a custom Lambda step that writes dataset paths, code S3 URIs, and hyperparameters to a DynamoDB table after each pipeline run.

Answer

Enable CloudTrail on the account so that all SageMaker API calls are logged, then query the logs to reconstruct the model lineage.

Answer

Store each pipeline execution's inputs and outputs manually as tags on the deployed endpoint configuration.