In this appendix we illustrate the process of instatiating the Gradient Boosting algorithm of Table 4.8 to a particular differentiable loss function.
We need to solve:
This is done in two steps:
- Step 1: given c, solve for Tm = T (x; pm):
- Step 2: given Tm, solve for c:
Solving Equation (B.2) for robust loss functions L(y,) such as the absolute loss, Huber loss, binomial deviance, etc., ...