private int ClosestCentroidToCentroid(Centroid c)
{
double lowestDist = Double.MaxValue;
int lowestIndex = 0;
for (int i = 0; i < _centroids.Length; i++)
{
var tmpDist = GetDistance(c, _centroids[i]);
if (tmpDist < lowestDist && _centroids[i].Length > 1)
{
lowestDist = tmpDist;
lowestIndex = i;
}
}
return lowestIndex;
}
private int ClosestPoint(Centroid c1, Centroid c2)
{
double lowestDist = GetDistance(c2.GetPoint(0), c1);
int lowestIndex = 0;
for (int i = 1; i < c2.Length; i++)
{
var tmpDist = GetDistance(c2.GetPoint(i), c1);
if (tmpDist < lowestDist)
{
lowestDist = tmpDist;
lowestIndex = i;
}
}
return lowestIndex;
}
void Initialize()
{
double distortionBeforeUpdate = 0; // distortion measure before
// updating centroids
double distortionAfterUpdate = 0; // distortion measure after update
// centroids
// design a 1-vector Codebook
_centroids = new Centroid[1];
// then initialize it with (0, 0) coordinates
double[] origin = new double[_dimension];
_centroids[0] = new Centroid(origin);
// initially, all training points will belong to 1 single cell
for (int i = 0; i < _points.Length; i++)
{
_centroids[0].Add(_points[i], 0);
}
// calls update to set the initial codevector as the average of all
// points
_centroids[0].Update();
// Iteration 1: repeat splitting step and K-means until required number
// of codewords is reached
while (_centroids.Length < _codebookSize)
{
// split codevectors by a binary splitting method
Split();
// group training points to centroids closest to them
GroupPtoC();
// Iteration 2: perform K-means algorithm
do
{
for (int i = 0; i < _centroids.Length; i++)
{
distortionBeforeUpdate += _centroids[i].Distortion;
_centroids[i].Update();
}
// regroup
GroupPtoC();
for (int i = 0; i < _centroids.Length; i++)
{
distortionAfterUpdate += _centroids[i].Distortion;
}
} while (Math.Abs(distortionAfterUpdate - distortionBeforeUpdate) < MinDistortion);
}
}
private void Split()
{
// "Centroids length now becomes " + centroids.length + 2
Centroid[] temp = new Centroid[_centroids.Length * 2];
double[][] tCo = new double[2][];
for (int i = 0; i < temp.Length; i += 2)
{
tCo[0] = new double[_dimension];
tCo[1] = new double[_dimension];
for (int j = 0; j < _dimension; j++)
{
tCo[0][j] = _centroids[i / 2][j] * (1 + SplitValue);
tCo[1][j] = _centroids[i / 2][j] * (1 - SplitValue);
}
temp[i] = new Centroid(tCo[0]);
temp[i + 1] = new Centroid(tCo[1]);
}
// replace old centroids array with new one
_centroids = temp;
}
private double GetDistance(Point point, Centroid tC)
{
double distance = 0;
for (int i = 0; i < _dimension; i++)
{
var temp = point[i] - tC[i];
distance += temp * temp;
}
distance = Math.Sqrt(distance);
return distance;
}
