Hyperparameter
A configuration variable (e.g., learning rate, tree depth) set before model training that influences learning behavior and performance.
Definition
Critical knobs that shape model complexity, convergence speed, and generalization. Governance demands cataloguing hyperparameters in experiment- tracking systems, applying consistent tuning protocols, and locking hyperparameters for production models to ensure reproducibility. Periodic reviews may adjust hyperparameters to address data-drift or new performance targets.
Real-World Example
A data-science team tunes a random-forest classifier’s tree-depth and minimum-sample-leaf parameters via grid search, logs the hyperparameter settings yielding best validation AUC, and seeds the production pipeline with those exact values—ensuring that the deployed model matches the reported performance.
.svg)
