ldaModel: Combine Model-based Recursive Partitioning with Linear...

In schiffner/locClass: Collection of Local Classification Methods

Description

Combine Model-Based Recursive Partitioning with Linear Discriminant Analysis.

Usage

ldaModel

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

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

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

predict.ldaModel(object, out = c("class", "posterior"), ...)

Arguments

object	An object of class "ldaModel" and "wlda", respectively.
weights	A vector of observation weights.
x	An object of class "wlda".
out	Should class labels or posterior probabilities be returned?
...	Further arguments.

Format

An object of class StatModel of length 1.

Details

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

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

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

Value

reweight: The re-weighted fitted "ldaModel" object.
deviance: The value of the deviance for Linear Discriminant Analysis extracted from object, i.e. the log-likelihood.
estfun: The empirical estimating (or score) function for Linear Discriminant Analysis, i.e. the derivatives of the log-likelihood with respect to the parameters, evaluated at the training data.
predict: Either a vector of predicted class labels 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

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 = ldaModel,
control = mob_control(objfun = deviance, minsplit = 200))

## predict posterior probabilities
pred <- predict(fit, newdata = grid, out = "posterior")
post <- do.call("rbind", pred)

image(x, x, matrix(as.numeric(post[,1]), length(x)), xlab = "x.1", ylab = "x.2")
contour(x, x, matrix(as.numeric(post[,1]), length(x)), levels = 0.5, 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) != as.numeric(data$y))

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