Question

A machine learning team uses SageMaker Pipelines to retrain a customer-churn model weekly. After training, an evaluation step computes the model's F1 score against a holdout set. The team wants the pipeline to register the newly trained model in the SageMaker Model Registry only if its F1 score exceeds 0.80; otherwise, the pipeline should stop without registering the model. Which approach implements this requirement within the pipeline?

Accepted Answer

Add a ConditionStep that reads the F1 score from the evaluation step's property file and, if the condition is met, executes a RegisterModel step.

Answer

Configure the training step to automatically register the model, then use a Lambda function outside the pipeline to delete it if the F1 score is below 0.80.

Answer

Set a CloudWatch alarm on the F1 metric emitted during training and trigger model registration through an EventBridge rule.

Answer

Use a CallbackStep to pause the pipeline and require a data scientist to manually approve registration based on the F1 score.