Module loss 

Loss functions for GBDT training

Provides objective functions with gradient and hessian computation:

• MseLoss: Mean Squared Error (standard, but sensitive to outliers)
• PseudoHuberLoss: Robust loss that transitions smoothly from L2 to L1
• BinaryLogLoss: Binary cross-entropy for binary classification
• MultiClassLogLoss: Softmax cross-entropy for multi-class classification

Also provides activation functions:

• sigmoid: Numerically stable sigmoid for binary classification
• softmax: Numerically stable softmax for multi-class classification

Structs

BinaryLogLoss

Binary Log Loss (Binary Cross-Entropy)

MseLoss

Mean Squared Error (L2) loss

MultiClassLogLoss

Multi-class LogLoss (Softmax Cross-Entropy)

PseudoHuberLoss

Pseudo-Huber Loss

Traits

LossFunction

Objective function for gradient boosting

Functions

sigmoid

Sigmoid function with numerical stability

softmax

Softmax function with numerical stability