Question

A team is running SageMaker Automatic Model Tuning on an XGBoost model. Last week they completed a tuning job with 40 training jobs. This week they want to launch a new tuning job that explores the SAME hyperparameter ranges on the SAME dataset, but they want the new job to reuse knowledge from the previous job's results to converge faster and avoid re-exploring poor regions. Which warm start configuration should they use?

Accepted Answer

Use warm start with the IDENTICAL_DATA_AND_ALGORITHM type, referencing the previous tuning job as the parent

Answer

Use warm start with the TRANSFER_LEARNING type, referencing the previous tuning job as the parent

Answer

Do not use warm start; instead increase the max number of parallel training jobs so the new job finishes faster

Answer

Use warm start with the TRANSFER_LEARNING type and expand the hyperparameter ranges to include new values