Certified Machine Learning ProfessionalDemo

Question

A Machine Learning Engineer has automated a model-retraining job in Databricks. Every scheduled run trains multiple candidate models using new sales data and logs all runs in MLflow. The objective is to select and register the highest-performing model at the end of each cycle to maintain optimal forecast accuracy.

Which approach meets this objective?

A Register the first model that has an evaluation metric better than the previously deployed model.
B Register the best model based on the primary evaluation metric.
C Register the model with the lowest training loss regardless of validation metrics.
D Register all models from the run since all of them are trained on newer data and hence will be more accurate.

Explanation

Candidate models should be compared using the defined primary evaluation metric, which measures the performance relevant to forecast accuracy. Registering the model with the best value on that metric selects the strongest validated candidate; training loss, run order, and data recency are not sufficient indicators of production-quality performance. Databricks documents using MLflow metrics to compare models and select versions for registration.

Learn more

Certified Machine Learning ProfessionalDemo

AstroTutor

Notes

Readiness

QuestionQ106

Community Discussion