scpp

PURPOSE

Spectral Clustering Projection Pursuit (SCPP) (divisive clustering is implemented in scppdc.m)

SYNOPSIS

function [optS, idx, spindex] = scpp(Data, pars, labels, colours)

DESCRIPTION

Spectral Clustering Projection Pursuit (SCPP) (divisive clustering is implemented in scppdc.m)
[OPTS, IDX, SPINDEX] = SCPP(X, PARS, LABELS, COLOURS)

 Returns:
    (OPTS): Linear subspace that minimises second eigenvalue of normalised Laplacian constructed
        from projected data
    (IDX): Binary cluster assignment {-1,1}
    (SPINDEX): Value of splitting index criterion 

 Inputs:
    (X): Data matrix
    (PARS): Structure containing all parameters of scpp() algorithm
    (LABELS): True clusters; only used for visualisation (optional)
    (COLOURS): Colormap matrix: only used for visualisation (optional)

CROSS-REFERENCE INFORMATION

This function calls:

fixLabels Enforces cluster labels to be in the range 1:K
ifelse Shorthand for ternary operator: if-then-else
isOctave Determines whether the environment is GNU Octave (returns TRUE) or MATLAB (returns FALSE)
pcacomp Returns the principal components of (X) specified in vector (index)
f_df_sc Function value and derivative of second smallest eigenvalue of normalised Laplacian
f_sc Second smallest eigenvalue of Normalised Laplacian and difference to 3rd smallest
microcluster Performs k-means with large K to estimate micro-clusters
reverse_assign Assigns observations to clusters based on the cluster assignment of the micro-clusters
schp Class implementing a linear projection subspace of arbitrary dimensions estimated through SCPP
scppNJW Clustering of projected data using Ng, Jordan and Weiss (2002) normalised spectral clustering algorithm
scpp_def_sigma Default scaling parameter (sigma) for Gaussian kernel used to estimate kernel/ similarity matrix in SCPP

This function is called by:

scppdc Spectral Clustering Projection Pursuit Divisive Clustering
schp Class implementing a linear projection subspace of arbitrary dimensions estimated through SCPP