Question

A data scientist is running a SageMaker Automatic Model Tuning job for an XGBoost model. To finish the tuning as fast as possible, they want to run many training jobs concurrently (high parallelism). They notice that when they set the strategy to a method that learns from prior results, increasing parallelism does not improve the quality of the hyperparameters chosen. Which explanation and adjustment is most accurate?

Accepted Answer

Bayesian optimization uses results from completed jobs to choose the next set of hyperparameters, so high parallelism reduces the information available for each new suggestion; use random search if maximizing concurrency is the priority.

Answer

Random search uses results from completed jobs, so increasing parallelism degrades it; switch to Bayesian optimization to run all jobs at once.

Answer

Grid search is the only strategy that benefits from parallelism; convert the tuning job to grid search to fully utilize concurrent jobs.

Answer

High parallelism never affects tuning quality; the issue is that the objective metric regex is not matching the training logs.