Q: 5
Scenario: A forecasting pipeline needs retraining on a larger dataset with a different distribution. Budget is limited, so the new tuning job must leverage previously saved high-performing hyperparameters, and must automatically stop if validation loss does not improve. Question- Which hyperparameter tuning job configuration should be used?. Options:
Options
Discussion
Makes sense to pick A here.
A. not D. TRANSFER_LEARNING is specifically for cases where the data distribution changes, which is what the scenario says. D's IDENTICAL_DATA_AND_ALGORITHM only works if the dataset hasn't changed, so that's a trap here.
Option A
A
I don’t think it’s D. A matches because the data distribution changed, so only TRANSFER_LEARNING warm start (option A) lets you reuse old hyperparams from a different dataset. Early stopping helps with the budget too. Pretty sure this is what AWS expects for scenarios like this, but chime in if you see a catch.
A
D
I think A. IDENTICAL_DATA_AND_ALGORITHM (D) is only for same dataset, but here the distribution has changed so transfer learning warm start makes sense.
A tbh, since D is tempting but that one only works when the data distribution doesn't change.
C/D? AMT Hyperband in C would be better for pure cost reduction, but the question wants to reuse prior hyperparams even with distribution change, so A makes sense. TRANSFER_LEARNING warm start fits that need, plus Early Stopping covers budget. Pretty sure it's A but open to other logic.
Be respectful. No spam.
