public static int SquaredEuclideanDistance(Point point, Centroid centroid)
{
//(A.x - B.x)^2 + (A.y - B.y)^2
return (point.X - centroid.X) * (point.X - centroid.X) +
(point.Y - centroid.Y) * (point.Y - centroid.Y) ;
}
public void Start(Boolean needVisualEterations, ref Int32 numberofIterations)
{
// Input:
// K = clustersCount - number of clusters
// {Point1, Point2, ... , PointN} - dataset of given points(vectors) {x1, x2, ... , xN}
// Output:
// {Centroid1, Centroid2, ... , CentroidK} - dataset of new cluster centroids {jm1, jm2, ... , jmK}
//
// Implementation:
// {Centroid1, Centroid2, ... , CentroidK} <-- SelectRandomSeeds({Point1, Point2, ... , PointN}, K)
List<Centroid> prevCentroids;
do
{
numberofIterations++;
prevCentroids = new List<Centroid>(centroids);
// Step 1 - Clear Clusters
for (Int32 k = 0; k < clustersCount; k++)
clusters[k].Clear();
// Step 2 - Find Closest Centroid
for (Int32 n = 0; n < pointsCount; n++)
{
var point = points[n];
Int32 j = K_Math.ArgMin(K_Math.SquaredEuclideanDistances(point, centroids));
clusters[j].Add(points[n]); // Wj = Wj U {Xn}
}
if (needVisualEterations)
{
graphics.Clear(Color.White);
DrawClusters();
}
//Step 3 - Update Centroids
for (Int32 k = 0; k < clustersCount; k++)
{
// 1
// jm[k] = -------- * ESum(X), where X C W[k];
// | W[k] |
Point sum = new Point(0, 0);
foreach (var point in clusters[k])
{
sum.X += point.X;
sum.Y += point.Y;
}
Int32 newX = sum.X /= clusters[k].Count;
Int32 newY = sum.Y /= clusters[k].Count;
centroids[k] = new Centroid(newX, newY, centroids[k].Color);
}
if (needVisualEterations)
{
DrawCentroids();
Thread.Sleep(100);
}
} while (!Centroid.AreEqualCentroids(prevCentroids, centroids));
if (!needVisualEterations)
{
graphics.Clear(Color.White);
DrawClusters();
}
Thread.Sleep(1000);
}