selectComponents: Select the Most Informative Components

In tensorBSS: Blind Source Separation Methods for Tensor-Valued Observations

Description

Takes an array of observations as an input and outputs a subset of the components having the most extreme kurtoses.

Usage

selectComponents(x, first = 2, last = 2)

Arguments

x	Numeric array of an order at least two. It is assumed that the last dimension corresponds to the sampling units.
first	Number of components with maximal kurtosis to be selected. Can equal zero but the total number of components selected must be at least two.
last	Number of components with minimal kurtosis to be selected. Can equal zero but the total number of components selected must be at least two.

Details

In independent component analysis (ICA) the components having the most extreme kurtoses are often thought to be also the most informative. With this viewpoint in mind the function selectComponents selects from x first components having the highest kurtosis and last components having the lowest kurtoses and outputs them as a standard data matrix for further analysis.

Value

Data matrix with rows corresponding to the observations and the columns correponding to the first + last selected components in decreasing order with respect to kurtosis. The names of the components in the output matrix correspond to the indices of the components in the original array x.

Author(s)

Joni Virta

Examples

library(ElemStatLearn)
x <- zip.train

rows <- which(x[, 1] == 0 | x[, 1] == 1)
x0 <- x[rows, 2:257]

x0 <- t(x0)
dim(x0) <- c(16, 16, 2199)

tfobi <- tFOBI(x0)
comp <- selectComponents(tfobi$S)
head(comp)

tensorBSS documentation built on May 29, 2017, 5:13 p.m.

Plot an Object of the Class tbss

tensorBSS index

Blind Source Separation Methods for Tensor-Valued...