Databricks Certified Machine Learning Professional — Question 82

A Machine Learning Engineer is using joblibspark and MLflowCallback to perform a distributed hyperparameter tuning experiment via Optuna. Assuming they have a single objective they are optimizing for, what will be the default sampler implemented by Optuna?

Answer options

• A. GridSampler
• B. RandomSampler
• C. NSGAIISampler
• D. TPESampler

Correct answer: D

Explanation

The default sampler in Optuna for optimizing a single objective is the TPESampler, which utilizes a tree-structured Parzen estimator for sampling. The GridSampler is not suitable for dynamic tuning, RandomSampler lacks the efficiency of the TPE approach, and NSGAIISampler is designed for multi-objective optimization, making them incorrect choices in this context.