rocsvm.path {rocsvm.path}

Fit the entire regularization path for ROC-Support Vector Machine (ROC-SVM)

Description

This algorithm computes the entire regularization path for the ROC-Support Vector Machine with a relatively low cost compared to quadratic programming problem.

Usage

rocsvm.path(x, y, rho = 1, kernel = poly.kernel, param.kernel = 1,
  prop = 0.5, lambda.min = 1e-05, eps = 1e-05, Nmoves = 500)

Arguments

x

The data matrix (n x p) with n rows (observations) on p variables (columns)

y

The {-1, 1} valued response variable.

rho

A positive constant

kernel

This is a user-defined function. Provided options are polynomial kernel; poly (the default, with parameter set to default to a linear kernel) and radial kernel; radial.

param.kernel

The parameter(s) for the kernel. For this radial kernel, the parameter is known in the fields as "gamma". For the polynomial kernel, it is the "degree"

prop

The proportion of large class corresponding a point of small class by speed-up tricks (the default is prop = 0.5). If you don't want to use the "speed-up tricks", then set prop to 1.

lambda.min

The smallest value of lambda for termination of the algorithm (the default is lambda.min = 1e-05).

eps

An adjustment computing errors

Nmoves

The maximum number of iterations the rocsvm.path algorithm

Value

A 'rocsvm.path' object is returned, for which there are lambda values and corresponding values of alpha for each data point.

Author(s)

Seung Jun Shin, Do Hyun Kim

See Also

rocsvm.get.solution, plot.rocsvm, rocsvm.intercept

Examples

library(rocsvm.path)
n <- 30
p <- 2
delta <- 1
set.seed(309)
y <- c(rep(1, n/2), rep(-1, n/2))
x <- matrix(0, n, p)
for (i in 1:n){
 if (y[i] == 1) {
 x[i,] <- rnorm(p, -delta, 1)
 } else {
 x[i,] <- rnorm(p, delta, 1)
  }
 }

rho = 1
kernel = radial.kernel
param.kernel  = 1/ncol(x)
prop = 0.1
obj <- rocsvm.path(x, y, rho, kernel, param.kernel, prop)