PearsonICA {PearsonICA}

Pearson-ICA Algorithm for Independent Component Analysis (ICA)

Description

The Pearson-ICA algorithm is a mutual information-based method for blind separation of statistically independent source signals. It has been shown that the minimization of mutual information leads to iterative use of score functions, i.e. derivatives of log densities. The Pearson system allows adaptive modeling of score functions. The flexibility of the Pearson system makes it possible to model a wide range of source distributions including asymmetric distributions. The algorithm is designed especially for problems with asymmetric sources but it works for symmetric sources as well.

Usage

PearsonICA(X, n.comp = 0, row.norm = FALSE, maxit = 200, tol = 1e-04, 
           border.base = c(2.6, 4), border.slope = c(0, 1), verbose = FALSE, 
		   w.init = NULL, na.rm = FALSE, whitening.only = FALSE, PCA.only = FALSE)

Arguments

X

input data. Each column contains one signal.

n.comp

number of components to be extracted.

row.norm

a logical value indicating whether rows of the data matrix 'X' should be standardized beforehand.

maxit

maximum number of iterations to perform

tol

a positive scalar giving the tolerance at which the un-mixing matrix is considered to have converged.

border.base

intercept terms for the tanh boundaries. See details.

border.slope

slope terms for the tanh boundaries. See details.

verbose

a logical value indicating the level of output as the algorithm runs.

w.init

initial un-mixing matrix of dimension (n.comp,n.comp). If NULL (default) then a matrix of normal r.v.'s is used.

na.rm

should the rows with missing values be removed.

whitening.only

perform only whitening.

PCA.only

perform only principal component analysis.

Details

The data matrix X is considered to be a linear combination of statistically independent components, i.e. X = SA where A is a linear mixing and matrix the columns of S contain the independent components of which at most one has Gaussian distribution. The goal of ICA is to find a matrix W such that the output Y = XW is an estimate of possibly scaled and permutated source matrix S.

In order to extract the independent components/sources we search for a demixing matrix W that minimizes the mutual information of the sources. The minimization of mutual information leads to iterative use of score functions, i.e. derivatives of log densities. Pearson-ICA uses the Pearson system to model the score functions of the output Y. The parameters of the Pearson system are estimated by method of moments. To speed up the algorithm, tanh nonlinearity is used when the distribution is far from Gaussian. The parameters 'border.base' and 'border.slope' define the boundaries of the tanh area in the skewness-kurtosis plane. See Figure 2 in (Karvanen, Eriksson and Koivunen, 2000) for an illustration.

Value

A list containing the following components

X

input data

whitemat

whitening matrix

W

estimated demixing matrix

A

estimated mixing matrix

S

separated (estimated) source signals

Xmu

component means

w.init

starting value of W

maxit

maximum number of iterations allowed

tol

convergence limit

it

number of iterations used

Warning

The definition of W is different from that of fastICA algorithm (version 1.1-8).

Note

The R code is based on the MATLAB code by Juha Karvanen, Jan Eriksson and Visa Koivunen. Parts of the R code and documentation are taken from the fastICA R package.

Author(s)

Juha Karvanen

References

Karvanen J., Koivunen V. 2002. Blind separation methods based on Pearson system and its extensions, Signal Processing 82(4), 663–673.

Karvanen J., Eriksson J.,Koivunen V. 2000, Pearson system based method for blind separation, Proceedings of Second International Workshop on Independent Component Analysis and Blind Signal Separation (ICA2000), Helsinki, Finland, 585–590.

See Also

PearsonICAdemo

Examples

S<-matrix(runif(5000),1000,5);
X<-S+S[,c(2,3,4,5,1)];
icaresults<-PearsonICA(X,verbose=TRUE)
print(icaresults$A)