Exam AWS Certified Machine Learning - Specialty topic 1 question 314 discussion

Reducing parallelism increases sequential execution time, making the job take longer overall.

AE https://docs.aws.amazon.com/sagemaker/latest/dg/automatic-model-tuning-considerations.html#automatic-model-tuning-num-hyperparameters

(ChatGPT) Option A (Hyperband): Efficiently utilizes computational resources. Reduces computation time by early stopping unpromising training jobs. Allows for a broader search of hyperparameter space within a shorter time. Option C (Lower MaxNumberOfTrainingJobs): Reduces the total number of training jobs. Directly decreases computation time. Helps stay within the small compute budget.

https://docs.aws.amazon.com/sagemaker/latest/dg/automatic-model-tuning-considerations.html#automatic-model-tuning-num-hyperparameters

Correct AE

A for sure C will reduce the tuning time because we are limiting the no. of Max Training jobs. E would be find with object is not reducing time

A: Yes - hyperband tuning early stops bad performing tuning jobs and reallocates resources to better performing ones B: No - Increase in hyperparams will increase time taken C: Yes - Limiting number of tuning jobs causes system not to run through entire list of tuning jobs reducing time D: No - grid search is computationally expensive and will take longer E: No - will increase time taken

A and D surely. Grid search to tune the hyperparameters.Grid search algorithm is more efficient than exhaustive search and will speed up tuning the hyperparameters.

Option A: Use the Hyperband tuning strategy. The Hyperband tuning strategy is a resource-efficient and time-saving approach for hyperparameter tuning. It works by running a set of hyperparameter configurations for a small number of training iterations and eliminating the poorly performing configurations early on. This strategy can significantly reduce the overall computation time compared to traditional methods like grid search or random search, especially for large hyperparameter spaces or time-consuming models like neural networks. Option E: Set a lower value for the MaxParallelTrainingJobs parameter. The MaxParallelTrainingJobs parameter in Amazon SageMaker specifies the maximum number of concurrent training jobs to be run in parallel during the hyperparameter tuning process. By setting a lower value for this parameter, the data scientist can limit the amount of computational resources used simultaneously, potentially reducing the overall computation time and cost.