Question

A data scientist is building a gradient-boosted model and plans to tune hyperparameters by comparing performance across many candidate configurations. They want an unbiased estimate of how the final chosen model will perform on future unseen data. They have a large dataset (2 million rows). How should they partition the data?

Accepted Answer

Split into train, validation, and test sets; tune hyperparameters using the validation set and report the final model's performance on the untouched test set.

Answer

Split into train and test only, tuning hyperparameters against the test set and reporting the best test score as the final estimate.

Answer

Use the entire dataset for both tuning and final evaluation to maximize the amount of data available.

Answer

Split into train and validation only, selecting the best model on validation and reporting that same validation score as the unbiased estimate.