Databricks Machine Learning Professional Question 34
Select 3A data science team is using Databricks to train multiple machine learning models for a production use case. They want to track and compare the performance of different experiments effectively, including hyperparameters, metrics, and model artifacts, to identify the best-performing model. Which of the following actions should they take to achieve this in Databricks?
- A
Log experiment runs using MLflow within Databricks and include hyperparameters, metrics, and artifacts.
- B
Manually record hyperparameters and metrics in a spreadsheet for each experiment run.
- C
Use MLflow's search capabilities to filter and sort experiment runs based on specific metrics.
- D
Enable automatic logging of metrics and parameters for supported ML libraries in Databricks.
- E
Delete older experiment runs to reduce clutter in the MLflow tracking server.
Show answer and explanation
Correct answers: A, C, D
Explanation
To effectively track and compare experiments in Databricks, the team should leverage MLflow's capabilities, including logging experiment metadata, using search and filtering features, and enabling automatic logging for supported libraries. These actions ensure that all relevant information is captured and accessible for model evaluation. Manual tracking or deleting runs is not recommended as it reduces efficiency or risks losing important data.
- A. Correct.
Logging experiment runs with MLflow is a core feature of Databricks and is essential for tracking hyperparameters, metrics, and artifacts in a structured and scalable way.
- B. Incorrect.
Manually recording hyperparameters and metrics is error-prone and does not leverage Databricks' built-in tools for experiment tracking. This action is not recommended.
- C. Correct.
MLflow's search capabilities allow users to filter and sort experiment runs, making it easier to compare models and identify the best one.
- D. Correct.
Enabling automatic logging in Databricks for supported ML libraries reduces manual effort and ensures consistent tracking of parameters and metrics.
- E. Incorrect.
Deleting older experiment runs is not necessary for effective experiment tracking, as MLflow provides filtering and sorting tools to manage clutter without losing valuable data.