/// <summary>
/// Split one cluster into two clusters according its mean
/// </summary>
/// <param name="iStart"></param>
/// <param name="iEnd"></param>
/// <param name="depth"></param>
/// <param name="maxDepth"></param>
private void SplitCluster(int iStart, int iEnd, int depth, int maxDepth)
{
if (iStart > iEnd)
{
return;
}
double mean, variance;
ComputeVariables(iStart, iEnd, out mean, out variance);
if (depth == maxDepth)
{
VQCluster c = new VQCluster(mean, variance, iStart, iEnd);
vqClusters.Add(c);
}
else
{
//Split the cluster into two clusters according mean value
int i;
for (i = iStart; i <= iEnd; i++)
{
//The following data will be greater than mean value, so we split it here
if (dataSet[i] > mean)
{
break;
}
}
SplitCluster(iStart, i - 1, depth + 1, maxDepth);
SplitCluster(i, iEnd, depth + 1, maxDepth);
}
}
/// <summary>
/// Build codebook according given data set
/// </summary>
/// <param name="vqSize"></param>
/// <returns></returns>
public double BuildCodebook(int vqSize)
{
if (vqSize > dataSetSize)
{
return(-1);
}
dataSet.Sort(0, dataSetSize);
//Set entire data as a single cluster, and then split it
double mean, var;
ComputeVariables(0, dataSetSize - 1, out mean, out var);
VQCluster c = new VQCluster(mean, var, 0, dataSetSize - 1);
vqClusters = new List <VQCluster>();
vqClusters.Add(c);
//Split clusters according its variance values
while (vqClusters.Count < vqSize)
{
int maxVarClusterId = MaxVarianceClusterId();
if (maxVarClusterId < 0)
{
break; // no more to split
}
//Split the cluster into two and remove the orginal one
SplitCluster(vqClusters[maxVarClusterId].iStart, vqClusters[maxVarClusterId].iEnd, 0, 1);
vqClusters.RemoveAt(maxVarClusterId);
}
//Adjust clusters according their mean values
AdjustCluster();
//Final codebook
vqSize = vqClusters.Count;
codebook = new double[vqSize];
double distortion = 0;
for (int i = 0; i < vqSize; i++)
{
codebook[i] = vqClusters[i].mean;
for (int j = vqClusters[i].iStart; j <= vqClusters[i].iEnd; j++)
{
double diff = dataSet[j] - codebook[i];
distortion += diff * diff;
}
}
distortion = Math.Sqrt(distortion / dataSetSize);
return(distortion);
}