Question

A data scientist runs several SageMaker training jobs manually over two weeks, adjusting learning rate, batch size, and dropout each time by editing a notebook cell. When asked to reproduce the configuration that produced the best validation F1 score, the scientist cannot reliably determine which hyperparameters generated that result because metrics and settings are scattered across notebook outputs. Which approach BEST addresses this problem going forward?

Accepted Answer

Integrate SageMaker Experiments so each training run is logged as a run within an experiment, capturing hyperparameters, input artifacts, and metrics for later comparison and reproduction.

Answer

Enable SageMaker Model Monitor on a development endpoint to capture the hyperparameters and metrics of each training job.

Answer

Store each trained model in the SageMaker Model Registry and rely on the version number to identify the hyperparameters used.

Answer

Write all hyperparameters and metrics to CloudWatch custom metrics and query them later with CloudWatch Logs Insights.