marytts.util.math

Class PCA

java.lang.Object

marytts.util.math.PCA

public class PCA
extends Object

Principal component analysis solve PCA using eigenvectors decomposion and singular value decomposition (SVD). Ref: Jonathon Shlens, "A tutorial on principal component analysis", (Dated: April 22, 2009; Version 3.01) http://www.snl.salk.edu/~shlens/pca.pdf

Author:

marcela

Constructor Summary

Constructors

Constructor and Description
PCA()

Method Summary

Modifier and Type	Method and Description
void	eigenPCA(Jama.Matrix data, boolean scale, boolean debug) Solving PCA using eigenvector decomposition
double[][]	getCovariance()
double[][]	getDataProjected(Jama.Matrix data, boolean debug)
double[]	getImportanceOfComponents()
double	getImportanceOfComponents(int numPC)
double[][]	getPC()
double[]	getV()
static void	main(String[] args)
void	principalComponentAnalysis(String fileName, boolean eigen, boolean scale) PCA
void	principalComponentAnalysis(Vector<Double> data, int rows, int cols, boolean eigen, boolean scale) perform principal component analysis
void	printImportanceOfComponents()
void	printPricipalComponents()
void	printPricipalComponents(String[] factors, int numPCA)
void	printPricipalComponents(Vector<String> factors)
void	svdPCA(Jama.Matrix data, boolean scale, boolean debug) Solving PCA using singular value decomposition (SVD) (more general solution)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

PCA

public PCA()

Method Detail

getCovariance

public double[][] getCovariance()

getV

public double[] getV()

getPC

public double[][] getPC()

getDataProjected

public double[][] getDataProjected(Jama.Matrix data,
                                   boolean debug)

printPricipalComponents

public void printPricipalComponents()

printPricipalComponents

public void printPricipalComponents(Vector<String> factors)

printPricipalComponents

public void printPricipalComponents(String[] factors,
                                    int numPCA)

Parameters:

factors - linguistic factors

numPCA - number of PC, between 1 and Max num PCs

printImportanceOfComponents

public void printImportanceOfComponents()

getImportanceOfComponents

public double[] getImportanceOfComponents()

getImportanceOfComponents

public double getImportanceOfComponents(int numPC)

principalComponentAnalysis

public void principalComponentAnalysis(Vector<Double> data,
                                       int rows,
                                       int cols,
                                       boolean eigen,
                                       boolean scale)

perform principal component analysis

Parameters:

data - a vector of doubles

rows - number of rows, trials or examples

cols - number of cols, dimensions or factors

eigen - if true use eigenvalues, if false use SVD (singular value decomposition)

scale - if true use znormalisation, if false just remove the mean from each dimension

eigenPCA

public void eigenPCA(Jama.Matrix data,
                     boolean scale,
                     boolean debug)

Solving PCA using eigenvector decomposition

Parameters:

data - Matrix with M rows corresponding to dimensions or factors and N columns corresponding to trials or examples

scale - if true : applying zscore normalisation if false: just removing the mean

debug - debug

svdPCA

public void svdPCA(Jama.Matrix data,
                   boolean scale,
                   boolean debug)

Solving PCA using singular value decomposition (SVD) (more general solution)

Parameters:

data - Matrix with M rows corresponding to dimensions or factors and N columns corresponding to trials or examples * @param scale if true : applying zscore normalisation if false: just removing the mean

scale - scale

debug - debug

principalComponentAnalysis

public void principalComponentAnalysis(String fileName,
                                       boolean eigen,
                                       boolean scale)

PCA

Parameters:

fileName - data one column per dimension or linguistic factor

eigen - if true use eigenvalues, if false use svd (recomended)

scale - if true use z-normalisation (recomended), if false substract off the mean for ecah dimension

main

public static void main(String[] args)
                 throws Exception

Throws:

Exception