/*
* public double ComputeVariance(double[] data)
* {
* double avg = data.Average();
* double sum = 0;
* foreach (double num in data)
* {
* sum += (num - avg)*(num - avg);
* }
* return sum / data.Length;
* }
*
* public double Gaussian(double num, double mean, double stddev) // 1-D gaussian
* {
* double res = (1 / (stddev * Math.Sqrt(2.0 * Math.PI))) *
* Math.Exp( (-1 * (num - mean) * (num - mean)) / (2 * stddev * stddev));
* return res;
* }
*/
private void btnInitialize_Click(object sender, EventArgs e)
{
PList.Clear();
Random rand = new Random();
double meanx0 = 100;
double meanx1 = 200;
double meanx2 = 325;
double meanx3 = 425;
double meany0 = 75;
double meany1 = 150;
double meany2 = 250;
double meany3 = 350;
double stddevx0 = 40;
double stddevx1 = 70;
double stddevx2 = 80;
double stddevx3 = 60;
double stddevy0 = 60;
double stddevy1 = 80;
double stddevy2 = 80;
double stddevy3 = 85;
int index = 0;
for (int i = 0; i < dataLength / 4; i++)
{
MyPoint pt = new MyPoint();
double rnum = rand.NextDouble();
if (rnum < 0.5)
{
pt.X = (int)(rand.NextDouble() * stddevx0 / 2 + meanx0);
}
else
{
pt.X = (int)(-1 * rand.NextDouble() * stddevx0 / 2 + meanx0);
}
if (rnum < 0.5)
{
pt.Y = (int)(rand.NextDouble() * stddevy0 / 2 + meany0);
}
else
{
pt.Y = (int)(-1 * rand.NextDouble() * stddevy0 / 2 + meany0);
}
index++;
PList.Add(pt);
}
for (int i = 0; i < dataLength / 4; i++)
{
MyPoint pt = new MyPoint();
double rnum = rand.NextDouble();
if (rnum < 0.5)
{
pt.X = (int)(rand.NextDouble() * stddevx1 / 2 + meanx1);
}
else
{
pt.X = (int)(-1 * rand.NextDouble() * stddevx1 / 2 + meanx1);
}
if (rnum < 0.5)
{
pt.Y = (int)(rand.NextDouble() * stddevy1 / 2 + meany1);
}
else
{
pt.Y = (int)(-1 * rand.NextDouble() * stddevy1 / 2 + meany1);
}
index++;
PList.Add(pt);
}
for (int i = 0; i < dataLength / 4; i++)
{
double rnum = rand.NextDouble();
MyPoint pt = new MyPoint();
if (rnum < 0.5)
{
pt.X = (int)(rand.NextDouble() * stddevx2 / 2 + meanx2);
}
else
{
pt.X = (int)(-1 * rand.NextDouble() * stddevx2 / 2 + meanx2);
}
if (rnum < 0.5)
{
pt.Y = (int)(rand.NextDouble() * stddevy2 / 2 + meany2);
}
else
{
pt.Y = (int)(-1 * rand.NextDouble() * stddevy2 / 2 + meany2);
}
PList.Add(pt);
index++;
}
for (int i = 0; i < dataLength / 4; i++)
{
double rnum = rand.NextDouble();
MyPoint pt = new MyPoint();
if (rnum < 0.5)
{
pt.X = (int)(rand.NextDouble() * stddevx3 / 2 + meanx3);
}
else
{
pt.X = (int)(-1 * rand.NextDouble() * stddevx3 / 2 + meanx3);
}
if (rnum < 0.5)
{
pt.Y = (int)(rand.NextDouble() * stddevy3 / 2 + meany3);
}
else
{
pt.Y = (int)(-1 * rand.NextDouble() * stddevy3 / 2 + meany3);
}
index++;
PList.Add(pt);
}
Visualization.DrawPoints(pic1, PList, 0.7);
MessageBox.Show("Data initialized for " + PList.Count + " points");
}
public double MInertia(Bitmap aoBmp)
{
double kdRes = 0.0;
double[] kdD = new double[this.k]; // Diameters for each cluster
double[ , ] kdP = new double[this.k, 4];
List <MyPoint>[] koList = new List <MyPoint> [k];
MyPoint[] koCenter = new MyPoint[k];
int kiC;
int x = 0, y = 0;
double kdPm;
MyPoint koPt;
for (int i = 0; i < k; i++)
{
kdD[i] = 0.0;
koList[i] = new List <MyPoint>( );
koCenter[i] = new MyPoint( );
koCenter[i].X = 0.0;
koCenter[i].Y = 0.0;
}
// Find all points in each cluster
for (int i = 0; i < X.Rows; i++)
{
// Gamma matrix has the probabilities for a data point for its membership in each cluster
kiC = 0;
kdPm = Gamma[i, 0];
for (int m = 0; m < k; m++)
{
if (Gamma[i, m] > kdPm)
{
kiC = m; // data i belongs to cluster m
kdPm = Gamma[i, m];
}
}
koPt = new MyPoint {
ClusterId = kiC, X = x, Y = y
};
koCenter[kiC].X += x;
koCenter[kiC].Y += y;
koList[kiC].Add(koPt);
x++;
if (x >= aoBmp.Width)
{
x = 0;
y++;
}
}
for (int i = 0; i < k; i++)
{
// Calculate the center of the cluster
koCenter[i].X /= koList[i].Count;
koCenter[i].Y /= koList[i].Count;
// Calculate the diameter of each cluster
for (int j = 0; j < koList[i].Count; j++)
{
kdD[i] += Math.Sqrt((koList[i][j].X - koCenter[i].X) * (koList[i][j].X - koCenter[i].X) +
(koList[i][j].Y - koCenter[i].Y) * (koList[i][j].Y - koCenter[i].Y));
}
// Normalize each diameter
kdD[i] /= koList[i].Count;
kdRes += kdD[i];
}
return(kdRes / this.ViK);
}
private void btnGMMND_Click(object sender, EventArgs e)
{
int k = int.Parse(txtNumClusters.Text); // number of clusters
int dim = 2; // number of dimensions for data
//int datSize = 20;
// read 2-D data
//FileInfo fi = new FileInfo("d:\\csharp2016\\clusterdata\\data2D.txt");
//FileInfo fi = new FileInfo("d:\\csharp2016\\clusterdata\\cdata1.txt");
//StreamReader sr = new StreamReader(fi.Open(FileMode.Open, FileAccess.Read));
//string sline = sr.ReadLine();
//int datSize = 1;
//while (sline != null)
//{
// datSize++;
// sline = sr.ReadLine();
//}
//sr.Close();
//sr = new StreamReader(fi.Open(FileMode.Open, FileAccess.Read));
//sline = sr.ReadLine();
//Matrix X = new Matrix(datSize, dim);
//int count = 0;
//while (sline != null)
//{
// string[] parts = sline.Split(new char[] { '\t', ' ' });
// X[count, 0] = double.Parse(parts[1].Trim());
// X[count, 1] = double.Parse(parts[2].Trim());
// count++;
// sline = sr.ReadLine();
//}
//sr.Close();
int datSize = 1000;
List <MyPoint> PList = InitializeData(datSize);
Mapack.Matrix X = new Matrix(datSize, dim);
for (int i = 0; i < PList.Count; i++)
{
X[i, 0] = PList[i].X;
X[i, 1] = PList[i].Y;
}
GMM_NDim gmmnd = new GMM_NDim(k, dim, X, dataSize);
gmmnd.ComputeGMM_ND();
// determine class membership i.e., which point belongs to which cluster
PList = new List <MyPoint>();
for (int i = 0; i < X.Rows; i++)
{
// Gamma matrix has the probabilities for a data point for its membership in each cluster
double[] probabs = new double[k];
int cnum = 0;
double maxprobab = gmmnd.Gamma[i, 0];
for (int m = 0; m < k; m++)
{
if (gmmnd.Gamma[i, m] > maxprobab)
{
cnum = m; // data i belongs to cluster m
maxprobab = gmmnd.Gamma[i, m];
}
}
MyPoint pt = new MyPoint {
ClusterId = cnum, X = X[i, 0], Y = X[i, 1]
};
PList.Add(pt);
}
MyImageProc.DrawClusters(pic1, PList, 1, k);
}
List <MyPoint> InitializeData(int datSize)
{
List <MyPoint> PList = new List <MyPoint>();
PList.Clear();
Random rand = new Random();
int dataLength = datSize; // number of data points
// create 4 distributions with different means and std devs
double meanx0 = 150;
double meanx1 = 180;
double meanx2 = 425;
double meanx3 = 475;
double meany0 = 175;
double meany1 = 250;
double meany2 = 300;
double meany3 = 320;
double stddevx0 = 100;
double stddevx1 = 90;
double stddevx2 = 220;
double stddevx3 = 260;
double stddevy0 = 50;
double stddevy1 = 80;
double stddevy2 = 180;
double stddevy3 = 195;
int index = 0;
for (int i = 0; i < dataLength / 4; i++)
{
MyPoint pt = new MyPoint();
double rnum = rand.NextDouble();
if (rnum < 0.5)
{
pt.X = rand.NextDouble() * stddevx0 / 2 + meanx0;
}
else
{
pt.X = -1 * rand.NextDouble() * stddevx0 / 2 + meanx0;
}
if (rnum < 0.5)
{
pt.Y = rand.NextDouble() * stddevy0 / 2 + meany0;
}
else
{
pt.Y = -1 * rand.NextDouble() * stddevy0 / 2 + meany0;
}
index++;
PList.Add(pt);
}
for (int i = 0; i < dataLength / 4; i++)
{
MyPoint pt = new MyPoint();
double rnum = rand.NextDouble();
if (rnum < 0.5)
{
pt.X = rand.NextDouble() * stddevx1 / 2 + meanx1;
}
else
{
pt.X = -1 * rand.NextDouble() * stddevx1 / 2 + meanx1;
}
if (rnum < 0.5)
{
pt.Y = rand.NextDouble() * stddevy1 / 2 + meany1;
}
else
{
pt.Y = -1 * rand.NextDouble() * stddevy1 / 2 + meany1;
}
index++;
PList.Add(pt);
}
for (int i = 0; i < dataLength / 4; i++)
{
double rnum = rand.NextDouble();
MyPoint pt = new MyPoint();
if (rnum < 0.5)
{
pt.X = rand.NextDouble() * stddevx2 / 2 + meanx2;
}
else
{
pt.X = -1 * rand.NextDouble() * stddevx2 / 2 + meanx2;
}
if (rnum < 0.5)
{
pt.Y = rand.NextDouble() * stddevy2 / 2 + meany2;
}
else
{
pt.Y = -1 * rand.NextDouble() * stddevy2 / 2 + meany2;
}
PList.Add(pt);
index++;
}
for (int i = 0; i < dataLength / 4; i++)
{
double rnum = rand.NextDouble();
MyPoint pt = new MyPoint();
if (rnum < 0.5)
{
pt.X = rand.NextDouble() * stddevx3 / 2 + meanx3;
}
else
{
pt.X = -1 * rand.NextDouble() * stddevx3 / 2 + meanx3;
}
if (rnum < 0.5)
{
pt.Y = rand.NextDouble() * stddevy3 / 2 + meany3;
}
else
{
pt.Y = -1 * rand.NextDouble() * stddevy3 / 2 + meany3;
}
PList.Add(pt);
index++;
}
MyImageProc.DrawClusters(pic1, PList, 1.0, 1);
return(PList);
}
//CREATING CLUSTERS AND SHOWING MEANS FOR EACH CLUSTER
private void btnCompute_Click(object sender, EventArgs e)
{
if (uploaded == true)
{
Stopwatch sw = new Stopwatch();
try
{
k = int.Parse(txtClusters.Text); //getting number of clusters
}
catch
{
MessageBox.Show("Wrong input.");
}
//sw.Start();
data = new DataPoints(picture, picture.Width, picture.Height);
PList = new List <MyPoint>(data.CreatePoints());
X = new Matrix(data_Size, dimenssions);
for (int i = 0; i < PList.Count; i++)
{
X[i, 0] = PList[i].Red;
X[i, 1] = PList[i].Green;
X[i, 2] = PList[i].Blue;
}
myGmm = new GMM_NDim(k, dimenssions, X);
sw.Start();
myGmm.ComputeGMM_ND();
sw.Stop();
MessageBox.Show("Time elapsed in ms: " + sw.ElapsedMilliseconds);
// determine class membership i.e., which point belongs to which cluster
PList = new List <MyPoint>();
for (int i = 0; i < X.Rows; i++) //looping throught all the rows = #points of Gamma -- each col = probability for each cluster
{
// Gamma matrix has the probabilities for a data point for its membership in each cluster
double[] probabs = new double[k];
int cnum = 0;
double maxprobab = myGmm.Gamma[i, 0];
for (int m = 0; m < k; m++) // Going throught the cols --- each col has probability for one cluster
{
if (myGmm.Gamma[i, m] > maxprobab) //checking for the cluster with more probability
{
cnum = m; // data i belongs to cluster m
maxprobab = myGmm.Gamma[i, m];
}
}
MyPoint pt = new MyPoint(X[i, 0], X[i, 1], X[i, 2], cnum); //X[i,0] = RED X[i,1] = GREEN X[i,2] = BLUE
PList.Add(pt);
}
computed = true;
showMeans();
}
else
{
MessageBox.Show("You need to upload an image first.");
}
}