Amazon SageMaker Automatic Model Tuning Now Supports Early Stopping of Training Jobs
Amazon SageMaker Automatic Model Tuning now supports early stopping of training jobs. With early stopping, training jobs will be automatically stopped during hyperparameter tuning when it becomes evident that they aren't likely to improve model accuracy. Early stopping will reduce your costs for hyperparameter tuning.
Amazon SageMaker Automatic Model Tuning now supports early stopping of training jobs. With early stopping, training jobs will be automatically stopped during hyperparameter tuning when it becomes evident that they aren't likely to improve model accuracy. Early stopping will reduce your costs for hyperparameter tuning.
During hyperparameter tuning, the number of possible configurations is exponential to the number of hyperparameters. The large number of variables means that it is possible that a new training job produces a less accurate model than what was previously achieved. When it becomes evident this is the case, you want to stop the training job and move on to another iteration. With this new enhancement, Amazon SageMaker will periodically compare an ongoing training job with previous training jobs to determine if it should be stopped. As long as you have not exceeded your tuning budget, a new job is automatically created to evaluate a different hyperparameter combination.
Early stopping of training jobs is now available in all AWS regions where Amazon SageMaker is available today. For more information, please read the feature documentation here.