Question

A machine learning team wants to automate their model build process so that every time new training code is pushed to their repository, a SageMaker Pipeline runs data processing, training, and evaluation steps automatically. They also need the pipeline definition itself to be version-controlled and deployed consistently across dev, test, and prod accounts. The team already uses AWS CodePipeline for their application code. What is the MOST appropriate way to implement this?

Accepted Answer

Use SageMaker Projects with the built-in MLOps template, which provisions CodePipeline/CodeBuild to build and deploy the SageMaker Pipeline definition from source control on each commit.

Answer

Manually run the SageMaker Pipeline from a notebook whenever code changes, and copy the pipeline definition JSON to each account by hand.

Answer

Schedule the SageMaker Pipeline to run every hour using an EventBridge rule so it always uses the latest code in S3.

Answer

Trigger a Lambda function on each commit that calls CreateTrainingJob directly, bypassing SageMaker Pipelines entirely.