private void btnKMeansClustering_Click(object sender, EventArgs e)
{
try
{
if (PList.Count == 0)
{
throw new Exception("No points data exists..");
}
int numClusters = int.Parse(txtClusters.Text);
List <ClusterCenterPoint> CList = null;
KMeans.DoKMeans(numClusters, ref PList, ref CList, 0.1, 1000, true); //true means do kplusplus
MyImageProc.DrawClusters(pic1, PList, 1.0, numClusters);
txtResult.Text = ComputeAndShowVarianceResults(numClusters);
}
catch (Exception ex)
{
MessageBox.Show(ex.Message);
}
}
private void btnKmeans_Click(object sender, EventArgs e)
{
try
{
if (PList.Count == 0)
{
throw new Exception("No points data exists..");
}
Stopwatch sw = new Stopwatch();
int numClusters = int.Parse(txtClusters.Text);
List <ClusterCenterPoint> CList = null;
sw.Start();
KMeans.DoKMeans(numClusters, ref PList, ref CList, 0.1, 1000);
sw.Stop();
MessageBox.Show("Time elapsed (ms): " + sw.ElapsedMilliseconds);
MyImageProc.DrawClusters(pic1, PList, 1.0, numClusters);
txtResult.Text = ComputeAndShowVarianceResults(numClusters);
}
catch (Exception ex)
{
MessageBox.Show(ex.Message);
}
}
public static int DoKMeansWithMinVariance(int numClusters, ref List <MyPoint> PList, ref List <ClusterCenterPoint> CList, double maxError, int maxIterations, bool minVariance)
{
double stddev = 0;
if (minVariance == true)
{
stddev = double.MaxValue;
}
else
{
stddev = double.MinValue;
}
List <MyPoint> PListBest = new List <MyPoint>();
List <ClusterCenterPoint> CListBest = new List <ClusterCenterPoint>();
int iter = 0;
for (int m = 0; m < 20; m++) //Pick best (most balanced) clustering after 20 clustering attempts
{
List <MyPoint> PListCopy = new List <MyPoint>();
foreach (MyPoint mp in PList)
{
PListCopy.Add((MyPoint)mp.Clone());
}
List <ClusterCenterPoint> CListCopy = null;
iter += KMeans.DoKMeans(numClusters, ref PListCopy, ref CListCopy, 0.01, 100, true); // true means do kmeansplusplus
// ----compute variance of cluster memberships
int[] CCount = new int[numClusters];
for (int i = 0; i < numClusters; i++)
{
CCount[i] = 0;
}
foreach (MyPoint mp in PListCopy)
{
CCount[mp.ClusterId] += 1;
}
double variance = 0;
for (int i = 0; i < numClusters; i++)
{
variance += (CCount[i] - (PListCopy.Count) / (double)numClusters) * (CCount[i] - (PListCopy.Count) / (double)numClusters);
}
double stddevCopy = Math.Sqrt(variance);
string out1 = "";
for (int n = 0; n < CCount.Length; n++)
{
out1 += "Cluster " + n.ToString() + " count = " + CCount[n].ToString() + "\n";
}
//MessageBox.Show("StdDev = " + stddevCopy.ToString() + " " + out1);
if (minVariance == true)
{
if (stddevCopy < stddev) // if it improves, copy data into best
{
stddev = stddevCopy;
PListBest.Clear();
foreach (MyPoint mp in PListCopy)
{
PListBest.Add((MyPoint)mp.Clone());
}
CListBest.Clear();
foreach (ClusterCenterPoint cp in CListCopy)
{
CListBest.Add((ClusterCenterPoint)cp.Clone());
}
}
}
else
{
if (stddevCopy > stddev) // if it improves, copy data into best
{
stddev = stddevCopy;
PListBest.Clear();
foreach (MyPoint mp in PListCopy)
{
PListBest.Add((MyPoint)mp.Clone());
}
CListBest.Clear();
foreach (ClusterCenterPoint cp in CListCopy)
{
CListBest.Add((ClusterCenterPoint)cp.Clone());
}
}
}
}
CList = CListBest;
PList = PListBest;
return(iter);
}
public static int DoKMeansWithMinVariance(int numClusters, ref List <MyPoint> PList, ref List <ClusterCenterPoint> CList, double maxError, int maxIterations, bool minVariance)
{
List <MyPoint> copy = new List <MyPoint>(PList);
object olock = new object();
double stddev = 0;
if (minVariance == true)
{
stddev = double.MaxValue;
}
else
{
stddev = double.MinValue;
}
List <MyPoint> PListBest = new List <MyPoint>();
List <ClusterCenterPoint> CListBest = new List <ClusterCenterPoint>();
int iter = 0;
Parallel.For(0, 20, (m) =>
//for (int m = 0; m < 20; m++) //Pick best (most balanced) clustering after 20 clustering attempts
{
List <MyPoint> PListCopy = new List <MyPoint>();
for (int i = 0; i < copy.Count; i++)
{
PListCopy.Add((MyPoint)copy[i].Clone());
}
//foreach (MyPoint mp in PList)
//PListCopy.Add((MyPoint)mp.Clone());
List <ClusterCenterPoint> CListCopy = null;
iter += KMeans.DoKMeans(numClusters, ref PListCopy, ref CListCopy, 0.01, 100, true); // true means do kmeansplusplus
// ----compute variance of cluster memberships
int[] CCount = new int[numClusters];
for (int i = 0; i < numClusters; i++)
{
CCount[i] = 0;
}
foreach (MyPoint mp in PListCopy)
{
CCount[mp.ClusterId] += 1;
}
double variance = 0;
for (int i = 0; i < numClusters; i++)
{
variance += (CCount[i] - (PListCopy.Count) / (double)numClusters) * (CCount[i] - (PListCopy.Count) / (double)numClusters);
}
double stddevCopy = Math.Sqrt(variance);
string out1 = "";
for (int n = 0; n < CCount.Length; n++)
{
out1 += "Cluster " + n.ToString() + " count = " + CCount[n].ToString() + "\n";
}
//MessageBox.Show("StdDev = " + stddevCopy.ToString() + " " + out1);
if (minVariance == true)
{
lock (olock) //Protect while updating stddev.
// Otherwise one thread X can obtain a X stddevcopy better than the actual one stop at the if validation
//then a new thread Y starts with Y stddevcopy better than actual one and X one, updates stddev and stops
//thread X starts again (does not need to pass if statement again) and updates stddevcopy which would be wrong since the best one was Y
{
if (stddevCopy < stddev) // if it improves, copy data into best
{
stddev = stddevCopy;
PListBest.Clear();
foreach (MyPoint mp in PListCopy)
{
PListBest.Add((MyPoint)mp.Clone());
}
CListBest.Clear();
foreach (ClusterCenterPoint cp in CListCopy)
{
CListBest.Add((ClusterCenterPoint)cp.Clone());
}
}
}
}
/*
* else
* {
* lock (olock)
* {
* if (stddevCopy > stddev) // if it improves, copy data into best
* {
* stddev = stddevCopy;
* PListBest.Clear();
* foreach (MyPoint mp in PListCopy)
* PListBest.Add((MyPoint)mp.Clone());
* CListBest.Clear();
* foreach (ClusterCenterPoint cp in CListCopy)
* CListBest.Add((ClusterCenterPoint)cp.Clone());
* }
* }
* }
*/
});
CList = CListBest;
PList = PListBest;
return(iter);
}
