Databricks Machine Learning Associate Question 571
Select 3You are working on a machine learning project in Databricks and need to optimize the hyperparameters of a Gradient Boosting model. You have decided to use hyperparameter search techniques available in MLflow. Which of the following statements are true about hyperparameter tuning using grid search, random search, or Bayesian optimization in Databricks?
- A
Grid search exhaustively evaluates all combinations of hyperparameters specified in the search space.
- B
Random search explores a fixed number of hyperparameter combinations randomly sampled from the search space.
- C
Bayesian optimization requires defining explicit ranges for every hyperparameter and evaluates all combinations in the range.
- D
Bayesian optimization iteratively selects the next set of hyperparameters to evaluate based on past evaluations.
- E
Grid search is typically computationally faster than Bayesian optimization for large hyperparameter spaces.
Show answer and explanation
Correct answers: A, B, D
Explanation
Hyperparameter tuning is a critical step in machine learning model development. Grid search evaluates all combinations, making it exhaustive but resource-intensive. Random search, on the other hand, samples hyperparameter combinations randomly, often yielding good results efficiently. Bayesian optimization builds a probabilistic model to iteratively select promising hyperparameters, making it efficient for large search spaces. Understanding the trade-offs of each method is essential when optimizing models in Databricks.
- A. Correct.
Correct. Grid search evaluates all possible combinations of hyperparameters within the specified search space, making it exhaustive but computationally expensive.
- B. Correct.
Correct. Random search selects a fixed number of hyperparameter combinations randomly, which can often yield good results with fewer evaluations compared to grid search.
- C. Incorrect.
Incorrect. Bayesian optimization does not evaluate all combinations in the range. Instead, it uses a probabilistic model to choose the most promising hyperparameters to evaluate.
- D. Correct.
Correct. Bayesian optimization utilizes past evaluations to decide the next set of hyperparameters, making it more efficient in identifying optimal settings.
- E. Incorrect.
Incorrect. Grid search is computationally slower than Bayesian optimization for large hyperparameter spaces because it evaluates all possible combinations exhaustively.