LPP Class Reference

#include <lpp.h>

Inheritance diagram for LPP:

Collaboration diagram for LPP:

Public Member Functions
void	setSpecificParameters (int k=12, double sigma=1.)
	Set specific parameters. More...
void	reduceDimensionality ()
	Reduce dimensionality. More...
Public Member Functions inherited from DimRedAlgorithm
	DimRedAlgorithm ()
	Empty constructor. More...
void	setInputData (Matrix2D< double > &X)
	Set input data. More...
void	setOutputDimensionality (size_t outputDim)
	Set output dimensionality. More...
const Matrix2D< double > &	getReducedData ()
	Get reduced data. More...

Public Attributes
int	k
double	sigma
Public Attributes inherited from DimRedAlgorithm
Matrix2D< double > *	X
	Pointer to input data. More...
size_t	outputDim
	Output dim. More...
Matrix2D< double >	Y
	Output data. More...
DimRedDistance2	distance
	Distance function. More...
FileName	fnMapping
	Save mapping. More...

Detailed Description

Class for making a LPP dimensionality reduction

Definition at line 37 of file lpp.h.

Member Function Documentation

reduceDimensionality()

void LPP::reduceDimensionality ( )

virtual

Reduce dimensionality.

Reduce dimensionality method based on the Locality Preserving Projections (LPP) algorithm. These are linear projective maps that arise by solving a variational problem that optimally preserves the neighborhood structure of the data set.

Implements DimRedAlgorithm.

Definition at line 37 of file lpp.cpp.

{
    // Compute the distance to the k nearest neighbors
    Matrix2D<double> D2;
    computeDistanceToNeighbours(*X, k, D2, distance, false);

    // Compute similarity matrix
    computeSimilarityMatrix(D2,sigma,true,true);

    // Compute graph laplacian
    Matrix2D<double> L;
    computeGraphLaplacian(D2,L);

    Matrix2D<double> DP, LP;
    matrixOperation_XtAX_symmetric(*X,D2,DP);
    matrixOperation_XtAX_symmetric(*X,L,LP);

    // Compute eigenvalues and eigenvectors resolving the generalized eigenvector problem
    Matrix2D<double> Peigvec, eigvector;
    Matrix1D<double> Deigval;
    generalizedEigs(LP, DP, Deigval, Peigvec);

    // Sort the eigenvalues in ascending order
    Matrix1D<int> idx;
    Deigval.indexSort(idx);

    // Sort the eigenvalues in descending order and get the smallest eigenvectors
    eigvector.resizeNoCopy(MAT_YSIZE(Peigvec),outputDim);
    for (size_t j=0; j<outputDim; ++j)
    {
        int idxj=VEC_ELEM(idx,j)-1;
        for (int i=0; i<MAT_YSIZE(Peigvec); ++i)
            MAT_ELEM(eigvector,i,j)=MAT_ELEM(Peigvec,i,idxj);
    }

    // Compute the result of the reduce dimensionality method
    Y=*X * eigvector;
    if (fnMapping!="")
        eigvector.write(fnMapping);
}

setSpecificParameters()

void LPP::setSpecificParameters	(	int	k = 12,
		double	sigma = 1.
	)

Set specific parameters.

Definition at line 28 of file lpp.cpp.

{
    this->k=k;
    this->sigma=sigma;
}

Member Data Documentation

k

int LPP::k

Definition at line 40 of file lpp.h.

sigma

double LPP::sigma

Definition at line 41 of file lpp.h.

---

The documentation for this class was generated from the following files:

• xmipp/libraries/dimred/lpp.h
• xmipp/libraries/dimred/lpp.cpp