Databricks Machine Learning Associate Question 95
Single answerYou are using the MLflow Client API to identify the best model from a set of runs logged in an experiment. The success metric for the model is 'rmse' (Root Mean Squared Error), where lower values indicate better performance. How would you programmatically identify the best run?
- A
Use the search_runs() method to filter runs by the 'rmse' metric and sort them in ascending order.
- B
Use the get_run() method to retrieve all runs and manually iterate through them to find the run with the lowest 'rmse' value.
- C
Use the list_artifacts() method to list all artifacts for each run and identify the best one based on the artifact metadata.
- D
Use the log_metric() method to overwrite the 'rmse' metric for all runs with the lowest value found in the experiment.
Show answer and explanation
Correct answer: A
Explanation
The search_runs() method in the MLflow Client API is the most efficient way to identify the best run when you need to query runs based on specific metrics like 'rmse'. It allows filtering and sorting to find the best-performing run programmatically. Other methods like get_run() or list_artifacts() are not designed for this purpose.
- A. Correct.
Correct. The search_runs() method allows you to query and filter runs based on specific conditions, such as the 'rmse' metric, and sort them to identify the best run efficiently.
- B. Incorrect.
Incorrect. While get_run() can retrieve individual run details, manually iterating through all runs is inefficient and not aligned with the best practices for using the MLflow Client API.
- C. Incorrect.
Incorrect. The list_artifacts() method is used to inspect run artifacts, not to evaluate or compare metrics like 'rmse'.
- D. Incorrect.
Incorrect. log_metric() is used to log metrics to a run, not to modify or evaluate metrics across multiple runs.