generate.loss.gr.huber: Generator for Huber Loss Function Gradient

In DPpack: Differentially Private Statistical Analysis and Machine Learning

Generator for Huber Loss Function Gradient

Description

This function generates and returns the Huber loss function gradient used for privacy-preserving SVM at the specified value of h in the form required by EmpiricalRiskMinimizationDP.CMS.

Usage

generate.loss.gr.huber(h)

Arguments

h	Positive real number for the Huber loss parameter. Lower values more closely approximate hinge loss. Higher values produce smoother Huber loss functions.

Value

Huber loss function gradient with parameter h in the form required by EmpiricalRiskMinimizationDP.CMS.

Examples

  h <- 1
  huber <- generate.loss.gr.huber(h)
  y.hat <- c(-.5, 1.2, -0.9)
  y <- c(-1, 1, -1)
  huber(y.hat,y)
  huber(y.hat, y, w=c(0.1, 0.5, 1)) # Weights observation-level loss gradient

DPpack documentation built on Oct. 20, 2024, 9:07 a.m.