getLocalDescription: Get the local description of an observation.

In tortnertuwien/lop: Local Projection Based Data Analysis

Description

Uses the k nearest neighbours of observation x. From those observations the most dense [k*alpha] observations are used to provide a projection onto the [k*alpha] oberservations. The orthogonal distance to this projection space and the score distance within this projection space, based on the covariance of the [k*alpha] observations are calculated for each observation from the dataset.

Usage

getLocalDescription(data, center, dist=NULL, k=20, alpha=0.5)

Arguments

data	A dataset, with n rows and p variables. Any matrix or data frame can be handed over.
center	An index of data, which defines the observation initiating the projection process.
dist	A distance matrix of data. If no distance matrix is provided, it will be calculated using dist(data). The parameter is used in order to make the calculation more efficient.
k=20	the number of nearest neighbours considered. k further influences the dimension of the projection space [k*alpha] at the same time.
alpha	The robustness parameter for the estimation of the covariance matrix. It further influences the dimension of the projection space.

Value

od	A vector of orthogonal distances of each observation with respect to the space spanned by the core observations.
cd	A vector of core distances of each observation with respect to the covariance structure defined by the core observations.
ecd	A vector of Euclidean core distances of each observation.
core	The indices of [k*alpha] observations used to span the projection space and estimate the covariance matrix.
knn	The k nearest neighbours of the center observation.
center	The observation initiating the local description.

Author(s)

Thomas Ortner ([e-mail])

Examples

library(lop)
locDesc <- getLocalDescription(glass$data, center=1, k=20, alpha=0.5)
pch <- rep(1, nrow(glass$data))
pch[locDesc$core] <- 4
plot(locDesc$od ~ locDesc$cd, col=glass$group, pch=pch)

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