A team wants a repeatable ML workflow that preprocesses data, trains a model, evaluates it, and registers the model only if its metrics pass a threshold. The whole flow should be defined as a directed graph and integrate natively with the model catalog. Which service is best?
Show answer and explanation
Answer: C
SageMaker Pipelines defines ML workflows as directed acyclic graphs of processing, training, evaluation, condition, and model registration steps. It integrates with the Model Registry, Experiments, and Feature Store, which gives lineage and governance across the workflow. Step Functions can call SageMaker APIs and coordinate general tasks, but it lacks ML-specific lineage and registration integration. EventBridge routes events, and CodePipeline automates application delivery stages.
Why the other options are wrong
- A. Step Functions is a general-purpose state machine service that can call SageMaker jobs. It works, but the team would build ML lineage and model registration logic itself instead of using native ML integrations.
- B. EventBridge is an event bus that routes events from sources to targets based on rules. It can trigger a workflow but does not define the processing, training, and evaluation steps.
- D. CodePipeline orchestrates software delivery stages such as build and deploy. It lacks native ML steps for processing, training, evaluation, and model registration.