Hyperparameter Tuning

Overview

Tuning squeezes the last 5-15% out of a model — but done carelessly it overfits the validation set and leaks preprocessing. The rules: tune the whole pipeline inside cross-validation, search smart (not grid), and keep a final untouched test set.

When to use

• A reasonable baseline exists and you want to improve it.
• You need to pick model complexity (depth, regularization, lr).

Strategy selection