AWS Certified Machine Learning – Specialty — Question 337

A data scientist uses Amazon SageMaker to perform hyperparameter tuning for a prototype machine leaming (ML) model. The data scientist's domain knowledge suggests that the hyperparameter is highly sensitive to changes.

The optimal value, x, is in the 0.5 < x < 1.0 range. The data scientist's domain knowledge suggests that the optimal value is close to 1.0.

The data scientist needs to find the optimal hyperparameter value with a minimum number of runs and with a high degree of consistent tuning conditions.

Which hyperparameter scaling type should the data scientist use to meet these requirements?

Answer options

• A. Auto scaling
• B. Linear scaling
• C. Logarithmic scaling
• D. Reverse logarithmic scaling

Correct answer: D

Explanation

Reverse logarithmic scaling is the correct choice because it allocates more search points near the upper bound of the range (close to 1.0), which matches the data scientist's domain knowledge. In contrast, logarithmic scaling concentrates search points near the lower bound (close to 0.5), and linear scaling distributes points uniformly across the entire range, making both less efficient for this scenario. Using reverse logarithmic scaling optimizes search efficiency and minimizes the required number of training runs.