jhplot.math.pca

Class PCA

java.lang.Object

jhplot.math.pca.PCA

public final class PCA
extends java.lang.Object

The class responsible mainly for preparing the PCA transformation parameters based on training data and executing the actual transformation on test data.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	PCA.TransformationType Type of the possible data transformation.

Constructor Summary

Constructors

Constructor and Description
PCA(Matrix data) Create the PCA transformation.
PCA(Matrix data, boolean center) Create the PCA transformation.
PCA(Matrix data, CovarianceMatrixEVDCalculator evdCalc) Create the PCA transformation.
PCA(Matrix data, CovarianceMatrixEVDCalculator evdCalc, boolean center) Create the PCA transformation

Method Summary

Modifier and Type	Method and Description
boolean	belongsToGeneratedSubspace(Matrix pt) Check if given point lies in PCA-generated subspace.
static Matrix	calculateCovarianceMatrix(Matrix data) Function for JUnit testing purposes only
double	getEigenvalue(int dimNo) Get selected eigenvalue
Matrix	getEigenvectorsMatrix()
int	getInputDimsNo() Get number of dimensions of the input vectors
int	getOutputDimsNo() Get number of dimensions of the output vectors
Matrix	transform(Matrix data, PCA.TransformationType type) Execute selected transformation on given data.

Methods inherited from class java.lang.Object

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

Constructor Detail

PCA

public PCA(Matrix data)

Create the PCA transformation. Use the popular SVD method for internal calculations

Parameters:

data - data matrix used to compute the PCA transformation. Rows of the matrix are the instances/samples, columns are dimensions. It is assumed that the matrix is already centered.

PCA

public PCA(Matrix data,
           boolean center)

Create the PCA transformation. Use the popular SVD method for internal calculations

Parameters:

data - data matrix used to compute the PCA transformation. Rows of the matrix are the instances/samples, columns are dimensions.

center - should the data matrix be centered before doing the calculations?

PCA

public PCA(Matrix data,
           CovarianceMatrixEVDCalculator evdCalc)

Create the PCA transformation.

Parameters:

data - data matrix used to compute the PCA transformation. Rows of the matrix are the instances/samples, columns are dimensions. It is assumed that the matrix is already centered.

evdCalc - method of computing eigenvalue decomposition of data's covariance matrix

PCA

public PCA(Matrix data,
           CovarianceMatrixEVDCalculator evdCalc,
           boolean center)

Create the PCA transformation

Parameters:

data - data matrix used to compute the PCA transformation. Rows of the matrix are the instances/samples, columns are dimensions.

evdCalc - method of computing eigenvalue decomposition of data's covariance matrix

center - should the data matrix be centered before doing the calculations?

Method Detail

getEigenvectorsMatrix

public Matrix getEigenvectorsMatrix()

Returns:

matrix where eigenvectors are placed in columns

getEigenvalue

public double getEigenvalue(int dimNo)

Get selected eigenvalue

Parameters:

dimNo - dimension number corresponding to given eigenvalue

getInputDimsNo

public int getInputDimsNo()

Get number of dimensions of the input vectors

getOutputDimsNo

public int getOutputDimsNo()

Get number of dimensions of the output vectors

transform

public Matrix transform(Matrix data,
                        PCA.TransformationType type)

Execute selected transformation on given data.

Parameters:

data - data to transform. Rows of the matrix are the instances/samples, columns are dimensions. If the original PCA data matrix was set to be centered, this matrix will also be centered using the same parameters.

type - transformation to apply

Returns:

transformed data

belongsToGeneratedSubspace

public boolean belongsToGeneratedSubspace(Matrix pt)

Check if given point lies in PCA-generated subspace. If it does not, it means that the point doesn't belong to the transformation domain i.e. it is an outlier.

Parameters:

pt - point. If the original PCA data matrix was set to be centered, this point will also be centered using the same parameters.

Returns:

true iff the point lies on all principal axes

calculateCovarianceMatrix

public static Matrix calculateCovarianceMatrix(Matrix data)

Function for JUnit testing purposes only