class CAAMDeformPCA

Performs Principal Component Analysis on a set of data vectors.

Inheritance:

CAAMDeformPCA - CAAMDeform - CAAMObject

Public Methods

void BackProject( const CDVector &param, CDVector &synth_obs ) const

Back projects a set of PCA model parameters into the original space.

int CalcNParam( const double retained_variance ) const

Calulates the needed number of parameters.

void ClearDataItems()

Deletes the data matrix

void Deform( const CDVector &params, CDVector &object ) const

Projects a set of PC scores to the original space.

void DoPCA( bool bWriteCoVarMarix )

Performs the principal component analysis.

void Dump( const char* szPath ) const

Writes eigen vectors and eigen values to matlab text files: pca_eigenvectors.m pca_eigenvalues.m pca_eigenvalues_org.m

inline const CDVector& EigenValues() const

Returns the eigen values in const vector form

inline CDVector& EigenValues()

Returns the eigen values in vector form

inline const CDVector& EigenValuesOrg() const

Returns the orignal eigen values in const vector form

inline CDMatrix& EigenVectors()

Returns the eigen vectors in matrix form

inline const CDMatrix& EigenVectors() const

Returns the eigen vectors in const matrix form

void Filter( CDVector &obs ) const

Projects an observation into the PCA space and back.

void FromFile( FILE* fh )

Reads a PCA object from file.

void InsertDataItem( const CDVector &v )

Inserts a data vector.

double MahalanobisDistance( const CDVector &params ) const

Returns the Mahalanobis distance of a set of PCA parameters.

const CDVector& MeanDataItem() const

Returns the sample mean.

const int NDataItems() const

Returns the number of data items this basis is based on

inline int NParameters() const

Returns the number of princal parameters

inline int NParametersOrg() const

Returns the number of princal parameters before any truncation

double ParameterWeight( const int i, bool asPercentage ) const

Returns the eigenvalue of the i-th parameter.

double ParameterWeightOrg( const int i, bool asPercentage ) const

Returns the eigenvalue of the i-th parameter.

void Project( const CDVector &obs, CDVector &param ) const

Projects an observation into the PCA space.

void ShuffleData()

Shuffle each dimension in all data items.

void ToFile( FILE* fh ) const

Writes a PCA object to file.

int TruncateParallel()

Truncate eigenvectors and eigenvalues using parallel analysis.

int TruncateVar( const double retained_variance )

Truncate eigenvectors and eigenvalues to retain 'retained_variance'.

void UseIdentityTransformation()

Makes the object use an identity basis instead of PCA basis.

Inherited from CAAMDeform:

Protected Fields

bool m_bValid

Inherited from CAAMObject:

Documentation

Performs Principal Component Analysis on a set of data vectors. The PCA basis can then be used for e.g. shape deformation.

inline CDVector& EigenValues()

Returns the eigen values in vector form

inline const CDVector& EigenValues() const

Returns the eigen values in const vector form

inline const CDVector& EigenValuesOrg() const

Returns the orignal eigen values in const vector form

inline CDMatrix& EigenVectors()

Returns the eigen vectors in matrix form

inline const CDMatrix& EigenVectors() const

Returns the eigen vectors in const matrix form

inline int NParameters() const

Returns the number of princal parameters

inline int NParametersOrg() const

Returns the number of princal parameters before any truncation

void ClearDataItems()

Deletes the data matrix

const int NDataItems() const

Returns the number of data items this basis is based on

double ParameterWeight( const int i, bool asPercentage ) const

Returns the eigenvalue of the i-th parameter in absolute numbers or as percentage.

Returns:

The eigenvalue.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

See Also:

ParameterWeightOrg

double ParameterWeightOrg( const int i, bool asPercentage ) const

Returns the eigenvalue of the i-th parameter in absolute numbers or as percentage.

Returns:

The eigenvalue.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

See Also:

ParameterWeightOrg

void InsertDataItem( const CDVector &v )

Inserts a data vector.

Parameters:

v - A data vector.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

void UseIdentityTransformation()

Makes the object use an identity basis instead of PCA basis, i.e. essentially a by-pass of this object.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-30-2002

void DoPCA( bool bWriteCoVarMarix )

Performs the principal component analysis on the data items. Uses the Eckhart-Young theorem if necessary for reduced memory and computational requirements.

Parameters:

bWriteCoVarMarix - If true the covariance matrix is written to the current dir. NOTE: Only in the case of more data items than dimensions (samples).

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

const CDVector& MeanDataItem() const

Returns the mean of all sample vectors.

Returns:

A mean vector.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

void ToFile( FILE* fh ) const

Writes a PCA object to a binary file.

Parameters:

fh - File handle to binary file open for writing.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-18-2000

See Also:

FromFile

void FromFile( FILE* fh )

Reads a PCA object from a binary file.

Parameters:

fh - File handle to binary file open for reading.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-18-2000

See Also:

ToFile

void Dump( const char* szPath ) const

Debug method that provides a human readable file dump of the object data.

Parameters:

szPath - Path including terminating backslash where the data is dumped.

Returns:

Nothing. @throws VisError

Author:

Mikkel B. Stegmann

Version:

10-18-2000

void Deform( const CDVector &params, CDVector &object ) const

Projects a set of PC scores to the original space.

Parameters:

params - PC scores, i.e. the model paramerisation.
object - Resulting projection into the original space, e.g. a shape.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

10-18-2000

double MahalanobisDistance( const CDVector &params ) const

Returns the Mahalanobis distance of a set of PCA parameters.

Parameters:

params - A set of PCA parameters.

Returns:

The Mahalanobis distance.

Author:

Mikkel B. Stegmann

Version:

6-11-2000

void ShuffleData()

Shuffle each dimension in all data items over all observations.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

7-17-2002

int TruncateParallel()

Truncate eigenvectors and eigenvalues using parallel analysis.

Also known as Horn's parallel analysis or Humphrey-Ilgen parallel analysis.

Assumes the data items have not been cleared.

Returns:

The number of parameters in the truncated basis.

Author:

Mikkel B. Stegmann

Version:

7-15-2002

See Also:

TruncateVar

int TruncateVar( const double retained_variance )

Truncate eigenvectors and eigenvalues to retain 'retained_variance'.

I.e. if retained_variance = .95 this function calculates how many eigenvectors we need to explain 95% of the total variation in the PCA training set.

Parameters:

retained_variance - Amount of variance to retain [0<x<1].

Returns:

The number of parameters in the truncated basis.

Author:

Mikkel B. Stegmann

Version:

6-11-2000

int CalcNParam( const double retained_variance ) const

Calulates the needed number of parameters to retain 'retained_variance'. I.e. if retained_variance = .95 then this function calculates how many eigenvectors we need to explain 95% of the total variation in the PCA training set.

Assumes that no eigenvalue cutoff has been done prior to the call.

Parameters:

retained_variance - Amount of variance to retain [0<x<1].

Returns:

The number of model parameters.

Author:

Mikkel B. Stegmann

Version:

6-5-2000

void Project( const CDVector &obs, CDVector &param ) const

Projects an observation into the PCA space.

Notice: Costly, due to the non-cached transpose of the eigenvectors.

Parameters:

obs - Input observation.
param - Output model parameters of the (possibly truncated) basis.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

3-6-2003

See Also:

BackProject, Filter

void BackProject( const CDVector &param, CDVector &synth_obs ) const

Back projects a set of PCA model parameters into the original space.

Parameters:

param - Input PCA model parameters.
synth_obs - Synthesized output observation.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

3-6-2003

See Also:

Project, Filter

void Filter( CDVector &obs ) const

Projects an observation into the PCA space and back.

Notice: Costly, due to the non-cached transpose of the eigenvectors in the project call.

Parameters:

obs - Input observation.

Returns:

Nothing.

Author:

Mikkel B. Stegmann

Version:

3-6-2003

See Also:

Project, BackProject

This class has no child classes.

Author:

Mikkel B. Stegmann

Version:

10-17-2000

See Also:

CAAMDeform

Alphabetic index Hierarchy of classes