A team set their SageMaker hyperparameter tuning job to stop based on MaxNumberOfTrainingJobs. They want tuning to halt automatically when the tuning algorithm determines that no configuration is likely to improve the objective metric by more than 1% over the current best job. Which tuning completion criterion achieves this behavior?

CompleteOnConvergence is the correct criterion because it specifically stops the tuning job when the algorithm determines that further training jobs are unlikely to yield significant improvements (e.g., less than 1%) to the objective metric. This optimizes resource usage by preventing unnecessary training. MaxRuntimeInSeconds stops the job after a specified duration, regardless of convergence. TargetObjectiveMetricValue stops the job once a predefined objective metric value is reached, not when improvement plateaus. MaxNumberOfTrainingJobsNotImproving is not a standard SageMaker hyperparameter tuning completion criterion.