Free Amazon MLS-C01 Actual Exam Questions - Question 1 Discussion
A data scientist is trying to improve the accuracy of a neural network classification model. The data
scientist wants to run a large hyperparameter tuning job in Amazon SageMaker.
However, previous smaller tuning jobs on the same model often ran for several weeks. The ML
specialist wants to reduce the computation time required to run the tuning job.
Which actions will MOST reduce the computation time for the hyperparameter tuning job? (Select
TWO.)
Maybe A and E, since Hyperband prunes early and more parallel jobs speed up runtime.
Maybe A and C, since Hyperband cuts early and fewer total jobs means less time.
A/E? Hyperband really speeds things up by cutting runs early, and upping MaxParallelTrainingJobs lets you run more trials at once, which should reduce wall-clock time significantly.
A/C makes sense, but I’d add that increasing MaxParallelTrainingJobs (not lowering it) speeds up the job by running more trials simultaneously. Grid search (D) usually takes longer, so that’s a no.
A definitely makes sense since Hyperband prunes bad runs early. On top of that, C is a solid choice because lowering MaxNumberOfTrainingJobs limits the total number of trials, which directly cuts down time. Increasing hyperparameters (B) or using grid search (D) usually lengthens tuning, so those don’t help here. Lowering MaxParallelTrainingJobs (E) actually slows things down since fewer jobs run at once, so that’s not ideal.
It’s A for sure since Hyperband stops bad trials early, saving tons of time. Also, I’d go with increasing MaxParallelTrainingJobs instead of lowering it because running more jobs at once speeds things up. So rather than C or E, I think A and just cranking parallel jobs is best to cut down total tuning time without missing good combos. Grid search (D) would only slow things down, and adding more hyperparameters (B) makes tuning longer, not shorter.
Option A and C. Hyperband cuts down unpromising jobs early, and lowering max training jobs limits total runs. Both reduce total time.