LocalDiscrimination: Local Discrimination based discrimination model

In tortnertuwien/lop: Local Projection Based Data Analysis

Description

Uses class-specific cores to compute local proections. The core distances and the orthogonal distance of each local projection are used to compute an LDA model. The models are then aggregated to an overall model.

Usage

localDiscrimination(data, class, k, dist=NULL)

Arguments

data	A data matrix containing the training data for the local projection based discrimination model. The observations are stored row-wise.
class	The class of observations, stored as a factor variable.
k	A tung parameter determining the degree of locality to be used for the local projections. It can be set to any natural number between 3 and the smallest group size-1. A good starting point is 50% of the smallest group size.

Value

core	The indices of observations used for the core.
group	The class of the observation initializing the local projection.
projection	A projection matrix used to project obersvations onto the core space. Computed using an SVD.
vf	Variable Filter. Variables with no variation within the core observations are removed as they cannot be re-scaled.
location	Estimation of location.
scale	Estimation of scatter.
core.location	Core based estimation of location.
core.scale	Core based estimation of scatter.
frequency	The frequency of the core occuring.
quality	A class specific description of the quality of the LDA model.
cv.posterior	Matrix of oosterior probabilities for each observation for each class computey by cross validation.
lda	The local LDA model of class "lda"
df.core	The centered and scaled core representation of the training observations.

Author(s)

Thomas Ortner ([e-mail])

References

Ortner, T., Hoffmann, I., Filzmoser, P., Zaharieva, M., Breiteneder, C., and Brodinova, S., ”Multigroup discrimination based on weighted local projections. Submitted to Journal of Computational and Graphical Statistics.

See Also

See Also predict.localDiscrimination, plot.localDiscrimination

Examples

library(lop)

x <- rrcovHD::olitos
data <- x[,1:25]
cl <- as.factor( x[,26] )
ld <- localDiscrimination(data, cl, k=7)
p0 <- predict(ld, data)
table(p0$class, cl)

lpplot(p0$weights, cl, combination=c(3,2))
lpplot(p0$weights, cl)

tortnertuwien/lop documentation built on May 30, 2019, 8:27 a.m.