How does the Huber loss change residual treatment compared with squared error for large mistakes?
Classic MLLevel: IntermediateRobust regression updates
How does the Huber loss change residual treatment compared with squared error for large mistakes?
Hints
- Compare slopes for small and large residuals.
- Robustness comes from limiting large-error influence.
Answer
It is quadratic near zero and grows linearly beyond its transition scale. Large residuals therefore have less rapidly growing influence, while small errors retain a smooth squared-error regime.
Source and licence: scikit-learn 1.7.2 — sgd · BSD-3-Clause · answer: Quantoros editorial