/// <summary>
/// /////PCA Generation//////////
/// </summary>
public void pca()
{
for (int i = 0; i < height * width; i++)
{
float sum = 0;
for (int j = 0; j < trainData; j++)
{
sum += C[i, j];
}
mean[i] = (float)(sum / trainData);
}
//-------------deviation from mean----------------------
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
dC[i, j] = (C[i, j] - mean[i]);
}
}
//--------------multiply transpose(dC)*dC---------------
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
tdC[j, i] = dC[i, j];
}
}
for (int i = 0; i < trainData; i++)
{
for (int j = 0; j < trainData; j++)
{
float sum = 0;
for (int k = 0; k < height * width; k++)
{
sum += tdC[i, k] * dC[k, j];
}
L[i, j] = sum / (trainData);
}
}
//-------------------calculate Eigenvector of L----------------//
MWArray c = null;
Eutilityclass Ev = new Eutilityclass();
c = Ev.Eigenvtr1((MWNumericArray)L);
MWNumericArray D = (MWNumericArray)c;
Array pA = D.ToVector(MWArrayComponent.Real);
float[] doubleArray1 = (float[])pA;
int icount = 0;
for (int i = 0; i < trainData; i++)
{
for (int j = 0; j < trainData; j++)
{
V[j, i] = doubleArray1[icount];
sV[j, trainData - i - 1] = doubleArray1[icount];
icount++;
}
}
savp("realEig.txt", (float[, ])sV.Clone());
/*
* //Eigen value
* c = Ev.Eigenvlu1((MWNumericArray)L);
* float[] eigenval = new float[L.GetLength(0)];
* D = (MWNumericArray)c;
* pA = D.ToVector(MWArrayComponent.Real);
* doubleArray1 = (float[])pA;
* icount = 0;
* int count = 0;
* for (int i = 0; i < L.GetLength(0); i++)
* {
* for (int j = 0; j < L.GetLength(1); j++)
* {
* if (j == i)
* {
* eigenval[count] = Math.Abs(doubleArray1[icount]);
* count++;
* }
* icount++;
* }
* }
* Vd = (float[])eigenval;
* //----------------------V=dC*V----------------------------
*/
float[] sm = new float[trainData];
float[] ssm = new float[trainData];
temp = matmul(ref dC, ref V);
stemp = matmul(ref dC, ref sV);
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
sm[j] += temp[i, j] * temp[i, j];
ssm[j] += stemp[i, j] * stemp[i, j];
}
}
//Improving classification by dividing by length (i think mormalization)
/*
* for (int j = 0; j < trainData; j++)
* {
*
* sm[j] = (float)Math.Pow(sm[j], .5);
* ssm[j] = (float)Math.Pow(ssm[j], .5);
*
* }
* */
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
temp[i, j] = temp[i, j] / (float)Math.Pow(sm[j], .5);
stemp[i, j] = stemp[i, j] / (float)Math.Pow(ssm[j], .5);
}
}
// obtaining the value of 'r'=tDc*V
/*
* float summ = 0;
*
* for (int i = 0; i < trainData; i++)
* {
* for (int j = 0; j < trainData; j++)
* {
*
* for (int k = 0; k < height * width; k++)
* {
* summ += tdC[i, k] * stemp[k, j];
*
* }
* rc[i, j] = summ;
* summ = 0;
* }
* }
*
*/
rc = matmul(ref tdC, ref stemp);
}
public void execute()
{
double[,] nm;
int m_b;// = noOfSuspects - 1;
double[,] Phi_w = new double[height * width, noOfImages];
int[,] t;
double[,] one;
double[,] mc;
double[,] res;
double[,] Phi_b;
int person = 0;
double[] eigenval;
double[,] eigenvtr;
double[] eigenval_sbt;
double[,] eigenvtr_sbt;
int[] index;
double[] aa;
double[] bb;
double[] tmp;
double[,] D_b;
double[,] mU_Sw_U;
double[] evl;
double[,] evr;
double[,] A;
double[] D_w;
//int m_b;
//int Sb_t;
//static int l=height*width;
//static int b=
for (int i = 0; i < noOfSuspects; i++)
{
breadth = imagePerClass[i];
t = new int[height * width, breadth];
one = new double[1, breadth];
for (int j = 0; j < breadth; j++)
{
one[0, j] = 1;
}
mc = new double[height * width, 1];
for (int j = 0; j < height * width; j++)
{
mc[j, 0] = mean_c[j, i];
for (int k = 0; k < imagePerClass[i]; k++)
{
t[j, k] = images[j, person + k];
}
}
nm = new double[mc.GetLength(0) * one.GetLength(0), mc.GetLength(1) * one.GetLength(1)];
k_t_p(ref mc, ref one, ref nm);
res = new double[nm.GetLength(0), nm.GetLength(1)];
for (int j = 0; j < nm.GetLength(0); j++)
{
for (int k = 0; k < nm.GetLength(1); k++)
{
res[j, k] = t[j, k] - nm[j, k];
}
}
for (int j = 0; j < height * width; j++)
{
for (int k = person; k < person + imagePerClass[i]; k++)
{
Phi_w[j, k] = res[j, k - person];
}
}
person = person + imagePerClass[i];
}
//int oo = 0;
for (int j = 0; j < Phi_w.GetLength(0); j++)
{
for (int k = 0; k < Phi_w.GetLength(1); k++)
{
Phi_w[j, k] = Phi_w[j, k] / Math.Sqrt(noOfImages);
}
}
one = new double[1, noOfSuspects];
for (int j = 0; j < noOfSuspects; j++)
{
one[0, j] = 1;
}
nm = new double[mean_i.GetLength(0) * one.GetLength(0), mean_c.GetLength(1) * one.GetLength(1)];
double[,] mi = new double[height * width, 1];
for (int j = 0; j < mi.GetLength(0); j++)
{
mi[j, 0] = mean_i[j];
}
k_t_p(ref mi, ref one, ref nm);
Phi_b = new double[mean_c.GetLength(0), mean_c.GetLength(1)];
for (int j = 0; j < mean_c.GetLength(0); j++)
{
for (int k = 0; k < mean_c.GetLength(1); k++)
{
Phi_b[j, k] = mean_c[j, k] - nm[j, k];
}
}
for (int i = 0; i < noOfSuspects; i++)
{
for (int j = 0; j < Phi_b.GetLength(0); j++)
{
Phi_b[j, i] = Phi_b[j, i] * Math.Sqrt((double)imagePerClass[i] / (double)noOfImages);
}
}
nm = new double[Phi_b.GetLength(1), Phi_b.GetLength(0)];
res = new double[nm.GetLength(0), Phi_b.GetLength(1)];
transpose(ref Phi_b, ref nm);
matrix_multiply(ref nm, ref Phi_b, ref res);
///calculation of Eigenvector of
MWArray c = null;
Eutilityclass Ev = new Eutilityclass();
c = Ev.Eigenvtr1((MWNumericArray)res);
MWNumericArray D = (MWNumericArray)c;
Array pA = D.ToVector(MWArrayComponent.Real);
double[] doubleArray1 = (double[])pA;
int icount = 0;
nm = new double[res.GetLength(0), res.GetLength(1)];//eigenvector
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
nm[j, i] = doubleArray1[icount];
icount++;
}
}
c = Ev.Eigenvlu1((MWNumericArray)res);
eigenval = new double[res.GetLength(0)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
icount = 0;
int count = 0;
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
if (j == i)
{
eigenval[count] = Math.Abs(doubleArray1[icount]);
count++;
}
icount++;
}
}
eigenval_sbt = new double[res.GetLength(1)];
eigenval_sbt = (double[])eigenval.Clone();
Array.Sort(eigenval);
index = new int[res.GetLength(0)];
for (int i = 0; i < eigenval_sbt.GetLength(0); i++)
{
index[i] = eigenval_sbt.GetLength(0) - 1 - Array.IndexOf(eigenval, eigenval[i]);
eigenval_sbt[i] = eigenval[eigenval_sbt.GetLength(0) - i - 1];
}
aa = new double[res.GetLength(0)];
eigenvtr_sbt = new double[res.GetLength(0), res.GetLength(1)];
count = 0;
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
eigenvtr_sbt[j, (int)index[i]] = nm[j, i];
}
aa[i] = eigenval_sbt[i] / eigenval_sbt[0];
if (aa[i] < thresh_eigval_sb)
{
count++;
}
}
bb = new double[count];
m_b = (int)((Math.Round(((double)(noOfSuspects - 1)) * remain_eigvec)));
count = 0;
//double temp;
for (int i = 0; i < eigenval_sbt.GetLength(0); i++)
{
if (aa[i] < thresh_eigval_sb)
{
bb[count] = i;
count++;
}
//temp=ei
}
//MessageBox.Show(""+m_b);
eigenvtr = new double[eigenvtr_sbt.GetLength(0), m_b];
eigenval = new double[m_b];
D_b = new double[m_b, m_b];
for (int j = 0; j < m_b; j++)
{
for (int i = 0; i < eigenvtr_sbt.GetLength(0); i++)
{
eigenvtr[i, j] = eigenvtr_sbt[i, j];
if (i == j)
{
D_b[i, j] = Math.Pow(eigenval_sbt[j], -1);
}
}
eigenval[j] = eigenval_sbt[j];
}
eigenvtr_sbt = new double[Phi_b.GetLength(0), eigenvtr.GetLength(1)];
matrix_multiply(ref Phi_b, ref eigenvtr, ref eigenvtr_sbt);
nm = new double[eigenvtr_sbt.GetLength(0), D_b.GetLength(0)]; ///Z
matrix_multiply(ref eigenvtr_sbt, ref D_b, ref (nm));
res = new double[nm.GetLength(1), nm.GetLength(0)]; //Z'
transpose(ref nm, ref res);
mc = new double[res.GetLength(0), Phi_w.GetLength(1)]; //mT
matrix_multiply(ref res, ref Phi_w, ref mc);
mU_Sw_U = new double[mc.GetLength(0), mc.GetLength(0)];
res = new double[mc.GetLength(1), mc.GetLength(0)];//mT'
transpose(ref mc, ref res);
matrix_multiply(ref mc, ref res, ref mU_Sw_U);
Phi_w = null;
Phi_b = null;
D_b = null;
eigenvtr = null;
eigenvtr_sbt = null;
icount = 0;
c = Ev.Eigenvlu1((MWNumericArray)mU_Sw_U);
evl = new double[mU_Sw_U.GetLength(1)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
for (int i = 0; i < mU_Sw_U.GetLength(0); i++)
{
for (int j = 0; j < mU_Sw_U.GetLength(1); j++)
{
if (i == j)
{
evl[j] = Math.Abs(doubleArray1[icount]);
}
icount++;
}
}
c = Ev.Eigenvtr1((MWNumericArray)mU_Sw_U);
evr = new double[mU_Sw_U.GetLength(0), mU_Sw_U.GetLength(1)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
count = 0;
for (int i = 0; i < (mU_Sw_U.GetLength(0)); i++)
{
for (int j = 0; j < (mU_Sw_U.GetLength(1)); j++)
{
evr[j, i] = (doubleArray1[count]);
count++;
}
}
tmp = new double[evl.GetLength(0)];
Array.Sort(evl);
index = new int[evl.GetLength(0)];
for (int i = 0; i < evl.GetLength(0); i++)
{
index[i] = Array.IndexOf(evl, evl[i]);
//tmp[i] = evl[evl.GetLength(0) - i-1];
}
//evl = (double[])tmp.Clone();
eigenvtr = new double[evr.GetLength(0), evr.GetLength(1)];//U_Vec
for (int i = 0; i < evr.GetLength(0); i++)
{
for (int j = 0; j < evr.GetLength(1); j++)
{
eigenvtr[j, (int)index[i]] = evr[j, i];
}
}
res = new double[nm.GetLength(0), eigenvtr.GetLength(1)];
matrix_multiply(ref nm, ref eigenvtr, ref res);
nm = null;
A = new double[res.GetLength(1), res.GetLength(0)];
transpose(ref res, ref A);
res = null;
evr = null;
eigenvtr = null;
D_w = new double[evl.GetLength(0)];
for (int i = 0; i < D_w.GetLength(0); i++)
{
//D_w[i]=eta_sw+evl[i];
D_w[i] = Math.Pow(eta_sw + evl[i], -(1.0 / 2.0));
}
res = new double[D_w.GetLength(0), D_w.GetLength(0)];
for (int i = 0; i < D_w.GetLength(0); i++)
{
res[i, i] = D_w[i];
}
D_b = null;
D_w = null;
DFLD_Trans = new double[res.GetLength(0), A.GetLength(1)];
matrix_multiply(ref res, ref A, ref DFLD_Trans);
res = null;
A = null;
y = new double[DFLD_Trans.GetLength(0), mean_c.GetLength(1)];
matrix_multiply(ref DFLD_Trans, ref mean_c, ref y);
//int aaaa;
}
public void execute()
{
double[,] nm;
int m_b;// = noOfSuspects - 1;
double[,] Phi_w = new double[height * width, noOfImages];
int[,] t;
double[,] one;
double[,] mc;
double[,] res;
double[,] Phi_b;
int person = 0;
double[] eigenval;
double[,] eigenvtr;
double[] eigenval_sbt;
double[,] eigenvtr_sbt;
int[] index;
double[] aa;
double[] bb;
double[] tmp;
double[,] D_b;
double[,] mU_Sw_U;
double[] evl;
double[,] evr;
double[,] A;
double[] D_w;
//int m_b;
//int Sb_t;
//static int l=height*width;
//static int b=
for (int i = 0; i < noOfSuspects; i++)
{
breadth = imagePerClass[i];
t = new int[height * width, breadth];
one = new double[1, breadth];
for (int j = 0; j < breadth; j++)
{
one[0, j] = 1;
}
mc = new double[height * width, 1];
for (int j = 0; j < height * width; j++)
{
mc[j, 0] = mean_c[j, i];
for (int k = 0; k < imagePerClass[i]; k++)
{
t[j, k] = images[j, person + k];
}
}
nm = new double[mc.GetLength(0) * one.GetLength(0), mc.GetLength(1) * one.GetLength(1)];
k_t_p(ref mc, ref one, ref nm);
res = new double[nm.GetLength(0), nm.GetLength(1)];
for (int j = 0; j < nm.GetLength(0); j++)
{
for (int k = 0; k < nm.GetLength(1); k++)
{
res[j, k] = t[j, k] - nm[j, k];
}
}
for (int j = 0; j < height * width; j++)
{
for (int k = person; k < person + imagePerClass[i]; k++)
{
Phi_w[j, k] = res[j, k - person];
}
}
person = person + imagePerClass[i];
}
//int oo = 0;
for (int j = 0; j < Phi_w.GetLength(0); j++)
{
for (int k = 0; k < Phi_w.GetLength(1); k++)
{
Phi_w[j, k] = Phi_w[j, k] / Math.Sqrt(noOfImages);
}
}
one = new double[1, noOfSuspects];
for (int j = 0; j < noOfSuspects; j++)
{
one[0, j] = 1;
}
nm = new double[mean_i.GetLength(0) * one.GetLength(0), mean_c.GetLength(1) * one.GetLength(1)];
double[,] mi = new double[height * width, 1];
for (int j = 0; j < mi.GetLength(0); j++)
{
mi[j, 0] = mean_i[j];
}
k_t_p(ref mi, ref one, ref nm);
Phi_b = new double[mean_c.GetLength(0), mean_c.GetLength(1)];
for (int j = 0; j < mean_c.GetLength(0); j++)
{
for (int k = 0; k < mean_c.GetLength(1); k++)
{
Phi_b[j, k] = mean_c[j, k] - nm[j, k];
}
}
for (int i = 0; i < noOfSuspects; i++)
{
for (int j = 0; j < Phi_b.GetLength(0); j++)
{
Phi_b[j, i] = Phi_b[j, i] * Math.Sqrt((double)imagePerClass[i] / (double)noOfImages);
}
}
nm = new double[Phi_b.GetLength(1), Phi_b.GetLength(0)];
res = new double[nm.GetLength(0), Phi_b.GetLength(1)];
transpose(ref Phi_b, ref nm);
matrix_multiply(ref nm, ref Phi_b, ref res);
///calculation of Eigenvector of
MWArray c = null;
Eutilityclass Ev = new Eutilityclass();
c = Ev.Eigenvtr1((MWNumericArray)res);
MWNumericArray D = (MWNumericArray)c;
Array pA = D.ToVector(MWArrayComponent.Real);
double[] doubleArray1 = (double[])pA;
int icount = 0;
nm = new double[res.GetLength(0), res.GetLength(1)];//eigenvector
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
nm[j, i] = doubleArray1[icount];
icount++;
}
}
c = Ev.Eigenvlu1((MWNumericArray)res);
eigenval = new double[res.GetLength(0)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
icount = 0;
int count = 0;
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
if (j == i)
{
eigenval[count] = Math.Abs(doubleArray1[icount]);
count++;
}
icount++;
}
}
eigenval_sbt = new double[res.GetLength(1)];
eigenval_sbt = (double[])eigenval.Clone();
Array.Sort(eigenval);
index = new int[res.GetLength(0)];
for (int i = 0; i < eigenval_sbt.GetLength(0); i++)
{
index[i] = eigenval_sbt.GetLength(0) - 1 - Array.IndexOf(eigenval, eigenval[i]);
eigenval_sbt[i] = eigenval[eigenval_sbt.GetLength(0) - i - 1];
}
aa = new double[res.GetLength(0)];
eigenvtr_sbt = new double[res.GetLength(0), res.GetLength(1)];
count = 0;
for (int i = 0; i < res.GetLength(0); i++)
{
for (int j = 0; j < res.GetLength(1); j++)
{
eigenvtr_sbt[j, (int)index[i]] = nm[j, i];
}
aa[i] = eigenval_sbt[i] / eigenval_sbt[0];
if (aa[i] < thresh_eigval_sb)
{
count++;
}
}
bb = new double[count];
m_b = (int)((Math.Round(((double)(noOfSuspects - 1)) * remain_eigvec)));
count = 0;
//double temp;
for (int i = 0; i < eigenval_sbt.GetLength(0); i++)
{
if (aa[i] < thresh_eigval_sb)
{
bb[count] = i;
count++;
}
//temp=ei
}
//MessageBox.Show(""+m_b);
eigenvtr = new double[eigenvtr_sbt.GetLength(0), m_b];
eigenval = new double[m_b];
D_b = new double[m_b, m_b];
for (int j = 0; j < m_b; j++)
{
for (int i = 0; i < eigenvtr_sbt.GetLength(0); i++)
{
eigenvtr[i, j] = eigenvtr_sbt[i, j];
if (i == j)
{
D_b[i, j] = Math.Pow(eigenval_sbt[j], -1);
}
}
eigenval[j] = eigenval_sbt[j];
}
eigenvtr_sbt = new double[Phi_b.GetLength(0), eigenvtr.GetLength(1)];
matrix_multiply(ref Phi_b, ref eigenvtr, ref eigenvtr_sbt);
nm = new double[eigenvtr_sbt.GetLength(0), D_b.GetLength(0)];///Z
matrix_multiply(ref eigenvtr_sbt, ref D_b, ref(nm));
res = new double[nm.GetLength(1), nm.GetLength(0)];//Z'
transpose(ref nm, ref res);
mc = new double[res.GetLength(0), Phi_w.GetLength(1)];//mT
matrix_multiply(ref res, ref Phi_w, ref mc);
mU_Sw_U = new double[mc.GetLength(0), mc.GetLength(0)];
res = new double[mc.GetLength(1), mc.GetLength(0)];//mT'
transpose(ref mc, ref res);
matrix_multiply(ref mc, ref res, ref mU_Sw_U);
Phi_w = null;
Phi_b = null;
D_b = null;
eigenvtr = null;
eigenvtr_sbt = null;
icount = 0;
c = Ev.Eigenvlu1((MWNumericArray)mU_Sw_U);
evl = new double[mU_Sw_U.GetLength(1)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
for (int i = 0; i < mU_Sw_U.GetLength(0); i++)
{
for (int j = 0; j < mU_Sw_U.GetLength(1); j++)
{
if (i == j)
{
evl[j] = Math.Abs(doubleArray1[icount]);
}
icount++;
}
}
c = Ev.Eigenvtr1((MWNumericArray)mU_Sw_U);
evr = new double[mU_Sw_U.GetLength(0), mU_Sw_U.GetLength(1)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (double[])pA;
count = 0;
for (int i = 0; i < (mU_Sw_U.GetLength(0)); i++)
{
for (int j = 0; j < (mU_Sw_U.GetLength(1)); j++)
{
evr[j, i] = (doubleArray1[count]);
count++;
}
}
tmp = new double[evl.GetLength(0)];
Array.Sort(evl);
index = new int[evl.GetLength(0)];
for (int i = 0; i < evl.GetLength(0); i++)
{
index[i] = Array.IndexOf(evl, evl[i]);
//tmp[i] = evl[evl.GetLength(0) - i-1];
}
//evl = (double[])tmp.Clone();
eigenvtr = new double[evr.GetLength(0), evr.GetLength(1)];//U_Vec
for (int i = 0; i < evr.GetLength(0); i++)
{
for (int j = 0; j < evr.GetLength(1); j++)
{
eigenvtr[j, (int)index[i]] = evr[j, i];
}
}
res = new double[nm.GetLength(0), eigenvtr.GetLength(1)];
matrix_multiply(ref nm, ref eigenvtr, ref res);
nm = null;
A = new double[res.GetLength(1), res.GetLength(0)];
transpose(ref res, ref A);
res = null;
evr = null;
eigenvtr = null;
D_w = new double[evl.GetLength(0)];
for (int i = 0; i < D_w.GetLength(0); i++)
{
//D_w[i]=eta_sw+evl[i];
D_w[i] = Math.Pow(eta_sw + evl[i], -(1.0 / 2.0));
}
res = new double[D_w.GetLength(0), D_w.GetLength(0)];
for (int i = 0; i < D_w.GetLength(0); i++)
{
res[i, i] = D_w[i];
}
D_b = null;
D_w = null;
DFLD_Trans = new double[res.GetLength(0), A.GetLength(1)];
matrix_multiply(ref res, ref A, ref DFLD_Trans);
res = null;
A = null;
y = new double[DFLD_Trans.GetLength(0), mean_c.GetLength(1)];
matrix_multiply(ref DFLD_Trans, ref mean_c, ref y);
//int aaaa;
}
/// <summary>
/// /////PCA Generation//////////
/// </summary>
public void pca()
{
for (int i = 0; i < height * width; i++)
{
float sum = 0;
for (int j = 0; j < trainData; j++)
{
sum += C[i, j];
}
mean[i] = (float)(sum / trainData);
}
//-------------deviation from mean----------------------
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
dC[i, j] = (C[i, j] - mean[i]);
}
}
//--------------multiply transpose(dC)*dC---------------
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
tdC[j, i] = dC[i, j];
}
}
for (int i = 0; i < trainData; i++)
{
for (int j = 0; j < trainData; j++)
{
float sum = 0;
for (int k = 0; k < height * width; k++)
{
sum += tdC[i, k] * dC[k, j];
}
L[i, j] = sum / (trainData);
}
}
//-------------------calculate Eigenvector of L----------------//
MWArray c = null;
Eutilityclass Ev = new Eutilityclass();
c = Ev.Eigenvtr1((MWNumericArray)L);
MWNumericArray D = (MWNumericArray)c;
Array pA = D.ToVector(MWArrayComponent.Real);
float[] doubleArray1 = (float[])pA;
int icount = 0;
for (int i = 0; i < trainData; i++)
{
for (int j = 0; j < trainData; j++)
{
V[j, i] = doubleArray1[icount];
sV[j, trainData - i - 1] = doubleArray1[icount];
icount++;
}
}
savp("realEig.txt", (float[,])sV.Clone());
/*
//Eigen value
c = Ev.Eigenvlu1((MWNumericArray)L);
float[] eigenval = new float[L.GetLength(0)];
D = (MWNumericArray)c;
pA = D.ToVector(MWArrayComponent.Real);
doubleArray1 = (float[])pA;
icount = 0;
int count = 0;
for (int i = 0; i < L.GetLength(0); i++)
{
for (int j = 0; j < L.GetLength(1); j++)
{
if (j == i)
{
eigenval[count] = Math.Abs(doubleArray1[icount]);
count++;
}
icount++;
}
}
Vd = (float[])eigenval;
//----------------------V=dC*V----------------------------
*/
float[] sm = new float[trainData];
float[] ssm = new float[trainData];
temp = matmul(ref dC, ref V);
stemp = matmul(ref dC, ref sV);
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
sm[j] += temp[i, j] * temp[i, j];
ssm[j] += stemp[i, j] * stemp[i, j];
}
}
//Improving classification by dividing by length (i think mormalization)
/*
for (int j = 0; j < trainData; j++)
{
sm[j] = (float)Math.Pow(sm[j], .5);
ssm[j] = (float)Math.Pow(ssm[j], .5);
}
* */
for (int i = 0; i < height * width; i++)
{
for (int j = 0; j < trainData; j++)
{
temp[i, j] = temp[i, j] / (float)Math.Pow(sm[j], .5);
stemp[i, j] = stemp[i, j] / (float)Math.Pow(ssm[j], .5);
}
}
// obtaining the value of 'r'=tDc*V
/*
float summ = 0;
for (int i = 0; i < trainData; i++)
{
for (int j = 0; j < trainData; j++)
{
for (int k = 0; k < height * width; k++)
{
summ += tdC[i, k] * stemp[k, j];
}
rc[i, j] = summ;
summ = 0;
}
}
*
*/
rc = matmul(ref tdC, ref stemp);
}
