Databricks Machine Learning Associate Question 452
Single answerYou are training multiple models in parallel and tracking their performance using MLflow in Databricks. After the training process completes, you need to programmatically identify the best run based on the lowest validation loss using the MLflow Client API. Which method would you use to achieve this?
- A
mlflow.search_runs() with an order_by argument to sort runs by validation loss
- B
mlflow.get_run() to retrieve the best run directly by specifying the metric to optimize
- C
mlflow.start_run() with a parameter to filter for the lowest validation loss
- D
mlflow.create_experiment() with a parameter to automatically track the best run
Show answer and explanation
Correct answer: A
Explanation
The MLflow Client API provides the mlflow.search_runs() method, which enables querying and sorting runs based on specific metrics, such as validation loss. By using the order_by argument, you can sort runs in ascending order of validation loss and programmatically identify the best run. Other methods like mlflow.get_run(), mlflow.start_run(), and mlflow.create_experiment() serve different purposes and are not suitable for this task.
- A. Correct.
Correct. mlflow.search_runs() allows you to query runs and use the order_by argument to sort the runs based on specific metrics, such as validation loss. You can retrieve the best run by sorting in ascending order of the validation loss.
- B. Incorrect.
Incorrect. mlflow.get_run() retrieves a specific run by its run ID but does not provide functionality to identify the best run based on a metric.
- C. Incorrect.
Incorrect. mlflow.start_run() is used to start a new run for experiment tracking but does not support filtering or sorting runs.
- D. Incorrect.
Incorrect. mlflow.create_experiment() is used to create a new experiment in MLflow and does not automatically track or identify the best run based on a metric.