Available on crate feature
alloc only.Expand description
Split criterion traits and XGBoost gain implementation.
This module defines how candidate splits are evaluated from histogram
gradient/hessian sums. The primary implementation is XGBoostGain,
which uses the exact gain formula from Chen & Guestrin (2016).
Structs§
- Split
Candidate - Result of evaluating potential splits across histogram bins.
- XGBoost
Gain - XGBoost-style split gain criterion.
Traits§
- Split
Criterion - Evaluates split quality from histogram gradient/hessian sums.
Functions§
- leaf_
weight - Compute the optimal leaf weight: -G / (H + lambda).