svmModel: Combine Model-based Recursive Partitioning with Support...

Description Usage Arguments Format Details Value References See Also Examples

Description

Combine Model-based Recursive Partitioning with Support Vector Machines.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
svmModel

\method{reweight}{svmModel}(object, weights, ...)

\method{deviance}{wsvm}(object, ...)

\method{estfun}{wsvm}(x, ...)

predict.svmModel(object, out = c("class", "posterior", "decision"), newdata,
  ...)

Arguments

object

An object of class "svmModel" and "wsvm", respectively.

weights

A vector of observation weights.

x

An object of class "wsvm".

out

Should class labels, posterior probabilities or decision values be returned?

newdata

A data.frame of cases to be classified.

...

Further arguments.

Format

An object of class StatModel of length 1.

Details

This page lists all ingredients to combine Support Vector Machines with Model-Based Recursive Partitioning (mob from package party). See the example for how to do that.

svmModel is an object of class StatModel-class implemented in package modeltools that provides an infra-structure for an unfitted wsvm model.

Moreover, methods for wsvm and svmModel objects for the generic functions reweight, deviance, estfun, and predict are provided.

Value

reweight: The re-weighted fitted "svmModel" object.
deviance: The value of the objective function extracted from object.
estfun: The empirical estimating (or score) function, i.e. the derivatives of the objective function with respect to the parameters, evaluated at the training data.
predict: Either a vector of predicted class labels, a matrix of decision values or a matrix of class posterior probabilities.

References

Zeileis, A., Hothorn, T. and Kornik, K. (2008), Model-based recursive partitioning. Journal of Computational and Graphical Statistics, 17(2) 492–514.

See Also

reweight, deviance, estfun, predict.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
library(benchData)

data <- vData(500)
x <- seq(0,1,0.05)
grid <- expand.grid(x.1 = x, x.2 = x)

fit <- mob(y ~ x.1 + x.2 | x.1 + x.2, data = data, model = svmModel, kernel = "linear", fitted = FALSE,
control = mob_control(objfun = deviance, minsplit = 200))

## predict decision values
dec <- predict(fit, newdata = grid, out = "decision")

image(x, x, matrix(dec, length(x)), xlab = "x.1", ylab = "x.2")
contour(x, x, matrix(dec, length(x)), levels = 0, add = TRUE)
points(data$x, pch = as.character(data$y))

## predict node membership
splits <- predict(fit, newdata = grid, type = "node")
contour(x, x, matrix(splits, length(x)), levels = min(splits):max(splits), add = TRUE, lty = 2)

## training error
mean(predict(fit) != data$y)

schiffner/locClass documentation built on May 29, 2019, 3:39 p.m.