Glossary term
Glossary term
Foundations
A hyperparameter in gradient boosting that controls overfitting. Shrinkage in gradient boosting is analogous to learning rate in gradient descent. Shrinkage is a decimal value between 0.0 and 1.0. A lower shrinkage value reduces overfitting more than a larger shrinkage value.
Created for this library
A risk modeling team uses shrinkage in its boosted-tree learning rate to produce a more conservative production model that generalizes well.
A research team tunes shrinkage in gradient boosting on a held-out set to balance training speed against generalization.
A retail forecasting team uses shrinkage in gradient boosting to keep individual tree contributions small and predictions stable.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License