Question 21
Domain 2: ML Model DevelopmentA machine learning engineer wants SageMaker Automatic Model Tuning to stop spending time on poorly performing training jobs and allocate more resources to promising hyperparameter configurations. Which tuning strategy best matches this goal?
Correct answer: D
Explanation
Use Hyperband when the goal is to aggressively prune weak training runs and focus resources on better-performing hyperparameter configurations. Bayesian, random, and grid search differ mainly in how they choose parameter values, not in early-stopping poor trials. — SageMaker Automatic Model Tuning strategies: Bayesian, random, grid, and Hyperband.
Why each option is right or wrong
A. Bayesian search, because it evaluates every combination in a fixed parameter lattice
Bayesian search selects promising values based on prior results; it does not evaluate a fixed lattice.
B. Random search, because it systematically enumerates the full hyperparameter space
Random search samples configurations stochastically rather than exhaustively covering all combinations.
C. Grid search, because it terminates weak jobs early and reallocates their budget
Grid search tests predefined combinations; early pruning of weak jobs is the distinguishing idea of Hyperband.
D. Hyperband, because it can stop underperforming jobs and concentrate effort on better candidates
Among SageMaker Automatic Model Tuning strategies, Hyperband is the method associated with allocating resources toward promising configurations while stopping poorly performing runs early. That directly matches the engineer's goal of avoiding wasted time on weak training jobs.