Loss

Description

Arguments

LossQuadratic

Quadratic differences between prediction and response. This loss
corresponds to the Gaussian distribution.

offset
Custom offset for initializing the model

LossAbsolute

Absolute differences between prediction and response. This loss
corresponds to the Laplace distribution.

offset
Custom offset for initializing the model

LossQuantile

Use this loss to boost arbitrary quantiles (robust regression).

offset
Custom offset for initializing the model
quantile
The quantile you like to boost

LossHuber

Huber loss with quadratic loss in \([d,d]\) and linear extrapolation outside.

offset
Custom offset for initializing the model
delta
Defining the interval in which the error is measured quadratically

LossBinomial

Binary classification loss that corresponds to the binomial distribution
with logit link. Labels are coded as 1 and 1.

offset
Custom offset for initializing the model

LossCustom

Custom loss by using custom C++ functions.

loss
R function to calculates the loss (vectorized)
gradient
R function to calculate the gradient
constant_initializer
R function to compute the optimal constant initialization

LossCustomCpp

Custom loss by using custom C++ functions.

loss_ptr
C++ pointer to a function that defines the loss
gradient_ptr
C++ pointer to a function to calculate the gradient
constant_initializer_ptr
C++ pointer to a function to compute the optimal constant
initialization
