Type: Package
Title: The Entire Solution Paths for ROC-SVM
Version: 0.1.0
Description: We develop the entire solution paths for ROC-SVM presented by Rakotomamonjy. The ROC-SVM solution path algorithm greatly facilitates the tuning procedure for regularization parameter, lambda in ROC-SVM by avoiding grid search algorithm which may be computationally too intensive. For more information on the ROC-SVM, see the report in the ROC Analysis in AI workshop(ROCAI-2004) : Hernàndez-Orallo, José, et al. (2004) <doi:10.1145/1046456.1046489>.
Imports: quadprog, svmpath
Depends: R (≥ 3.4.0)
License: GPL-2
Encoding: UTF-8
LazyData: true
RoxygenNote: 6.0.1
NeedsCompilation: no
Maintainer: Seung Jun Shin <sjshin@korea.ac.kr>
Packaged: 2018-10-07 00:50:22 UTC; KimDoHyun
Author: Seung Jun Shin [aut, cre], Do Hyun Kim [aut]
Repository: CRAN
Date/Publication: 2018-10-14 17:30:03 UTC

Plot the rocsvm.path, solution paths of ROC-SVM as a function of lambda

Description

produces a plot of the ROC-SVM lambda path.

Usage

## S3 method for class 'rocsvm'
plot(x, ...)

Arguments

x

The rocsvm path object

...

Generic compatibility

Value

The entire solution path of ROC-SVM solution as a function of lambda.

Author(s)

Seung Jun Shin, Do Hyun Kim

See Also

rocsvm.path

Examples

# The 'obj' comes from an example description of rocsvm.path()
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)
plot(obj)
# or plot.rocsvm(obj, lty = 2, lwd = 2, col = 2)

Compute the kernel matrix for ROC-SVM path

Description

Compute the kernel matrix for ROC-SVM path. This function comes from svmpath package by Trevor Hastie. If you want to know details of this function, refer the svmpath package.

Usage

poly.kernel(x, y = x, param.kernel = 1, ...)

Arguments

x

An n x p matrix of features

y

An m x p matrix of features

param.kernel

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

...

unused

Compute the kernel matrix for ROC-SVM path

Description

Compute the kernel matrix for ROC-SVM path. This function comes from svmpath package by Trevor Hastie. If you want to know details of this function, refer the svmpath package.

Usage

radial.kernel(x, y = x, param.kernel = 1/p, ...)

Arguments

x

An n x p matrix of features

y

An m x p matrix of features

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"

...

unused

Finding solutions fixed the regularization parameter of ROC-SVM.

Description

Computes solution alpha values from a fixed regularization parameter, lambda value for ROC-SVM path object.

Usage

rocsvm.get.solution(obj, lambda)

Arguments

obj

The rocsvm.path object

lambda

The regularization parameter that users want in ROC-SVM model.

Author(s)

Seung Jun Shin, Do Hyun Kim

See Also

rocsvm.path

Examples

# library(rocsvm.path)
# The 'obj' comes from an example description of rocsvm.path()

rocsvm.get.solution(obj, lambda = 1)

Finding an intercept fixed sensitivity or specificity for ROC-SVM

Description

Computes an intercept at a specific sensitivity or specificity level from the ROC-SVM model.

Usage

rocsvm.intercept(obj, lambda = 1, sensitivity = 0.5, specificity = 0.5)

Arguments

obj

The rocsvm.path object

lambda

The regularization parameter that users want in ROC-SVM model.

sensitivity

Sensitivity in ROC curve, which means True Positive Rate (TPR).

specificity

Specificity in ROC curve, which means True Negative Rate (TNR) = 1-FPR.

Author(s)

Seung Jun Shin, Do Hyun Kim

See Also

rocsvm.path

Examples

# library(rocsvm.path)
# The 'obj' comes from an example description of rocsvm.path()

rocsvm.intercept(obj, lambda = 1, sensitivity = 0.9, specificity = 0.1)

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)

Finding Lagrangian multipliers of ROC-SVM by Qudratic Programming

Description

Computes the Lagrangian multipliers(alpha), which are solutions of ROC-SVM using Quadratic Programming.

Usage

rocsvm.solve(K, lambda, rho = 1, eps = 1e-08)

Arguments

K

The kernelized matrix, i.e., K< .,. >.

lambda

The regularization parameter that users want in ROC-SVM model.

rho

A positive constant (default : 1)

eps

Adjustment computing errors (default : 1e-08)

Author(s)

Seung Jun Shin, Do Hyun Kim

See Also

rocsvm.path

Examples


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)
  }
 }

K <- radial.kernel(x,x)
rocsvm.solve(K, lambda = 1, rho = 1)