Databricks ML Associate Practice Question 3: MLflow Search Best Run
Databricks Certified Machine Learning Associate · Test 1 · question 3 of 10
An experiment holds about 200 runs from a regression tuning session, and each run logged the metric val_rmse. In a notebook, you need the ID of the single run with the best validation error so you can register its model.
Which call returns that run?
Show answer and explanation
Answer:
RMSE is an error metric, so lower is better. mlflow.search_runs returns matching runs as a pandas DataFrame. Sorting on metrics.val_rmse ASC puts the lowest error first, and max_results=1 keeps only that row. Its run_id column gives the run whose model you register.
Option B sorts descending, which returns the run with the worst validation RMSE. Descending order is right for metrics where higher is better, such as accuracy or R-squared.
Option C is not valid search syntax. filter_string compares a metric to a literal value (metrics.val_rmse < 5); it has no aggregate like min.
Option D returns the most recent run started by the current process, not the best one. It knows nothing about metric values.
Reference: Search Runs