PCA Class Reference

Performs Pricipal Component Analysis on samples with N variables. More...

List of all members.

Public Member Functions
	PCA (int nrvars)
virtual	~PCA ()
void	clearAllSamples ()
template<class IDXABL >
void	addSample (const IDXABL &sample)
bool	calculate ()
float	getEigenValue (int idx) const
template<class IDXABL >
void	getEigenVector (int idx, IDXABL &vec) const
void	setThreadWorker (Threads::WorkManager *)
Protected Member Functions
bool	tqli (float[], float[], int, ObjectSet< float > &)
void	tred2 (ObjectSet< float > &, int, float[], float[])
Protected Attributes
const int	nrvars
Array2DImpl< float >	covariancematrix
ObjectSet< TypeSet< float > >	samples
TypeSet< float >	samplesums
Threads::WorkManager *	threadworker
ObjectSet< SequentialTask >	tasks
float *	eigenvalues
int *	eigenvecindexes

---

Detailed Description

Performs Pricipal Component Analysis on samples with N variables.

Example of usage:

   //The samples will have three variables
   PCA pca( 3 );                

   //Samples can be added by any object that has a readable [] operator
   const float sample0[] = { 0, 1, 2 };
   pca.addSample( sample0 );

   const float sample1[] = { 4.343, 9.8, 2.72 };
   pca.addSample( sample1 );

   const float sample2[] = { 23.15, 210, -15 };
   pca.addSample( sample2 );

   const float sample3[] = { -0.36, 0.68, 3 };
   pca.addSample( sample3 );

   const float sample4[] = { 4.4, 9,6, 11 };
   pca.addSample( sample4 );

   TypeSet<float> sample5; sample5 += 34.1; sample5 += 8.37; sample5 += -44;
   pca.addSample( sample5 ); 

   pca.calculate();


   //Any object that has a writable [] operator can be used to fetch
   //the resulting vectors:
   TypeSet<float> eigenvec0(3,0);
   float eigenval0 = pca.getEigenValue(0);
   pca.getEigenVector( 0, eigenvec0 );

   float[3] eigenvec1;
   float eigenval1 = pca.getEigenValue(1);
   pca.getEigenVector( 1, eigenvec1 );

   float[3] eigenvec2;
   float eigenval2 = pca.getEigenValue(2);
   pca.getEigenVector( 2, eigenvec2 );

---

Constructor & Destructor Documentation

PCA::PCA

(

int

nrvars

)

Parameters:

nrvars

The number of variables that the samples have.

virtual PCA::~PCA

(

)

[virtual]

---

Member Function Documentation

template<class IDXABL >

void PCA::addSample

(

const IDXABL &

sample

)

[inline]

Adds a sample to the analysis.

Parameters:

sample

The sample that should be added. The sample can be of any type that have [] operators.

bool PCA::calculate

(

)

Computes the pca for all added samples.

void PCA::clearAllSamples

(

)

Removes all samples so a new analysis can be made (by adding new samples)

float PCA::getEigenValue

(

int

idx

)

const

Returns:

an eigenvalue.

Parameters:

idx

Determines which eigenvalue to return. The eigenvalues are sorted in descending order, so idx==0 gives the largest eigenvalue.

template<class IDXABL >

void PCA::getEigenVector	(	int	idx,
		IDXABL &	vec
	)			const [inline]

Returns the eigenvector corresponding to the eigenvalue idx.

Parameters:

	idx	Determines which eigenvector to return. The eigenvectors are sorted in order of descending eigenvalue, so idx==0 gives the eigenvector corresponding to the largest eigenvalue.
	vec	The object where the to store the eigenvector. Any object that has a writable [] operator can be used, for example float* and TypeSet<T>.

void PCA::setThreadWorker

(

Threads::WorkManager *

)

Enables multi-threaded calculation, which is beneficial when the number of samples and/or number of variables are large. The object does however work without setting the threadworkmanager.

Note:

threadworkmanager is managed (i.e. deleted) by caller, and it can be shared with many other objects (such as other PCAs).

bool PCA::tqli	(	float	[],
		float	[],
		int	,
		ObjectSet< float > &
	)			[protected]

void PCA::tred2	(	ObjectSet< float > &	,
		int	,
		float	[],
		float	[]
	)			[protected]

---

Member Data Documentation

Array2DImpl<float> PCA::covariancematrix [protected]

float* PCA::eigenvalues [protected]

The negation of the eigenval, to get the sorting right.

int* PCA::eigenvecindexes [protected]

const int PCA::nrvars [protected]

ObjectSet<TypeSet<float> > PCA::samples [protected]

TypeSet<float> PCA::samplesums [protected]

ObjectSet<SequentialTask> PCA::tasks [protected]

Threads::WorkManager* PCA::threadworker [protected]