XGBoost: Gradients and Hessians for Logistic Loss
Beginner
The problem, from first principles
To use cross-entropy loss inside boosting, XGBoost needs the first and second derivatives of the loss with respect to each sample's raw score. For logistic loss those derivatives have a short closed form.
From theory to code
Implement xgb_logistic_grad_hess(y, p), returning the gradient p - y and the hessian p (1 - p).
Constraints
pis a probability between 0 and 1.
Hints
Hint 1
The gradient is the prediction error, and the hessian is the variance of a Bernoulli with probability p.
Keywords
research-papersclassical-mlboostingxgboost
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.