source: cpp/frams/model/similarity/SVD/matrix_tools.cpp @ 1064

// This file is a part of Framsticks SDK.  http://www.framsticks.com/
// Copyright (C) 1999-2015  Maciej Komosinski and Szymon Ulatowski.
// See LICENSE.txt for details.


#include "matrix_tools.h"
#include "lapack.h"
#include <cstdlib>
#include <cmath>
#include <cstdio>
#include <stdlib.h> //malloc(), embarcadero
#include <math.h> //sqrt(), embarcadero


double *Create(int nSize)
{
        double *matrix = new double[nSize];

        for (int i = 0; i < nSize; i++)
        {
                matrix[i] = 0;
        }

        return matrix;
}

double *Multiply(double *&a, double *&b, int nrow, int ncol, double ncol2, double *&toDel, int delSize)
{
        double *c = Create(nrow * ncol2);
        int i = 0, j = 0, k = 0;

        for (i = 0; i < nrow; i++)
        {
                for (j = 0; j < ncol2; j++)
                {
                        for (k = 0; k < ncol; k++)
                                c[i * nrow + j] += a[i * nrow + k] * b[k * ncol + j];
                }
        }

        if (delSize != 0)
                delete[] toDel;
        return c;
}

double *Power(double *&array, int nrow, int ncol, double pow, double *&toDel, int delSize)
{
        double *m_Power = Create(nrow * ncol);
        if (pow == 2)
        {
                for (int i = 0; i < nrow; i++)
                {
                        for (int j = 0; j < ncol; j++)
                        {
                                m_Power[i * nrow + j] = array[i * nrow + j] * array[i * nrow + j];
                        }

                }
        }
        else
        {
                for (int i = 0; i < nrow; i++)
                {
                        for (int j = 0; j < ncol; j++)
                        {
                                m_Power[i * nrow + j] = sqrt(array[i * nrow + j]);
                        }

                }
        }

        if (delSize != 0)
                delete[] toDel;

        return m_Power;
}

void Print(double *&mat, int nelems)
{
        for (int i = 0; i < nelems; i++)
                printf("%6.2f ", mat[i]);
        printf("\n");

}

double *Transpose(double *&A, int arow, int acol, double *&toDel, int delSize)
{
        double *result = Create(acol * arow);

        for (int i = 0; i < acol; i++)
                for (int j = 0; j < arow; j++)
                {
                result[i * arow + j] = A[j * acol + i];
                }

        if (delSize != 0)
                delete[] toDel;
        
        return result;
}

//Weighted centring of a matrix.
//https://github.com/vegandevs/vegan/blob/master/src/goffactor.c
void wcentre(double *x, double *w, int *nr, int *nc)
{
        int i, j, ij;
        double sw, swx;

        for (i = 0, sw = 0.0; i < (*nr); i++)
                sw += w[i];

        for (j = 0; j < (*nc) ; j++)
        {
                for (i = 0, swx = 0.0, ij = (*nr)*j; i < (*nr); i++, ij++)
                {
                        swx += w[i] * x[ij];
                }
                swx /= sw;
                for (i = 0,  ij = (*nr)*j; i < (*nr); i++, ij++)
                {
                        x[ij] -= swx;
                        x[ij] *= sqrt(w[i]);
                }
        }
}

/** Computes the weighted MDS of the nSize x nSize distance matrix
                @param vdEigenvalues [OUT] Vector of doubles. On return holds the eigenvalues of the
                decomposed distance matrix (or rather, to be strict, of the matrix of scalar products
                created from the matrix of distances). The vector is assumed to be empty before the function call and
                all variance percentages are pushed at the end of it.
                @param nSize size of the matrix of distances.
                @param pDistances [IN] matrix of distances between parts.
                @param Coordinates [OUT] array of three dimensional coordinates obtained from SVD of pDistances matrix.
                @param weights [IN] vector of row weights.
                */
void MatrixTools::weightedMDS(std::vector<double> &vdEigenvalues, int nSize, double *pDistances, Pt3D *&Coordinates, double *weights)
{
        // compute the matrix D that is the parameter of SVD
        double *D = Create(nSize * nSize);
        D = Power(pDistances, nSize, nSize, 2.0, D, nSize);

        for (int i = 0; i < 2; i++)
        {
                wcentre(D, weights, &nSize, &nSize);
                D = Transpose(D, nSize, nSize, D, nSize);
        }
                
        for (int i = 0; i < nSize; i++)
                for (int j = 0; j < nSize; j++)
                {
                        D[i * nSize + j] *= -0.5;
                }

        double *Eigenvalues = Create(nSize);
        double *S = Create(nSize * nSize);

        // call the SVD function
        double *Vt = Create(nSize * nSize);
        size_t astep = nSize * sizeof(double);
        Lapack::JacobiSVD(D, astep, Eigenvalues, Vt, astep, nSize, nSize, nSize);

        double *W = Transpose(Vt, nSize, nSize, W, 0);

        delete[] D;
        delete[] Vt;
        
        // deweight
        double row_weight = 1;
        for (int i = 0; i < nSize; i++)
        {
                row_weight = weights[i];
                for (int j = 0; j < nSize; j++)
                {
                        W[i * nSize + j] /= sqrt(row_weight);
                }
        }

        for (int i = 0; i < nSize; i++)
                for (int j = 0; j < nSize; j++)
                {
                if (i == j)
                        S[i * nSize + j] = Eigenvalues[i];
                else
                        S[i * nSize + j] = 0;
                }

        // compute coordinates of points 
        double *sqS, *dCoordinates;
        sqS = Power(S, nSize, nSize, 0.5, S, nSize);
        dCoordinates = Multiply(W, sqS, nSize, nSize, nSize, W, nSize);
        delete[] sqS;

        for (int i = 0; i < nSize; i++)
        {
                // set coordinate from the SVD solution
                Coordinates[i].x = dCoordinates[i * nSize];
                Coordinates[i].y = dCoordinates[i * nSize + 1];
                if (nSize > 2)
                        Coordinates[i].z = dCoordinates[i * nSize + 2];
                else
                        Coordinates[i].z = 0;
        }

        // store the eigenvalues in the output vector
        for (int i = 0; i < nSize; i++)
        {
                double dElement = Eigenvalues[i];
                vdEigenvalues.push_back(dElement);
        }

        delete[] Eigenvalues;
        delete[] dCoordinates;
}