fastHICA (version 1.0.2)

basis_hica: Construction of the HICA basis

Description

This function builds the HICA tree up to a prespecified height providing the corresponding non-orthogonal bases.

Usage

basis_hica(X, maxlev = dim(X)[2] - 1, dim.subset = 512)

Arguments

X

Data matrix with nrow(X) observations and ncol(X) variables.

maxlev

The maximum level of the tree. This must be an integer between 1 and ncol(X)-1. The default value is set to ncol(X)-1.

dim.subset

The dimension of the subset used for the evaluation of the similarity index (i.e., distance correlation). If this it is greater than nrow(X) all the observations are used, unless a random subsample of dim.subset observations is used. The default value is set to 512.

Value

X

data matrix.

basis

a list with maxlev elements. The ith element of the list contains the basis matrix provided at level i of the tree. Each column of the basis matrix represent a basis element.

aggregation

a matrix with maxlev rows and 3 columns. At each row the first two columns contain the variable indeces merged at the corresponding level of the tree. In the third column the distance correlation of the two merged variables is recorded.

References

P. Secchi, S. Vantini, and P. Zanini (2014). Hierarchical Independent Component Analysis: a multi-resolution non-orthogonal data-driven basis. MOX-report 01/2014, Politecnico di Milano.

See Also

energy_hica, similarity_hica, extract_hica

Examples

## Not run:

##########################################################
# Example - Independent sources and overlapping loadings #
##########################################################

c1=c(0,0,0,0,1,1)
c2=c(1,1,1,1,0,0)
c3=c(1,1,0,0,0,0)

s1=runif(400,0,20)
s2=runif(400,0,20)
s3=runif(400,0,20)

# Here we generate the simulated dataset

X=s1%*%t(c1)+s2%*%t(c2)+s3%*%t(c3)+rnorm(6*400,0,1)

# Here we perform HICA on the simulated dataset

basis=basis_hica(X,5)

# Here we plot the 3 main components of HICA basis 
# (according to the energy criterium) for 4th level

energy=energy_hica(basis,6,5,plot=TRUE)
ex4=extract_hica(energy,3,4)
loa4=ex4$C

par( mfrow = c(3,1))
barplot(loa4[,1], ylim = c(-1, 1),main="HICA transform - Level 4",
ylab="1st component",xlab="Coordinate",names.arg=1:6,col="red",mgp=c(2.5,1,0))
barplot(loa4[,2], ylim = c(-1, 1),ylab="2nd component",
xlab="Coordinate",names.arg=1:6,col="green",mgp=c(2.5,1,0))
barplot(loa4[,3], ylim = c(-1, 1),ylab="3rd component",
xlab="Coordinate",names.arg=1:6,col="blue",mgp=c(2.5,1,0))

## End (Not run)