/// <summary>
/// Calculates a new centroid for every cluster based on the data of the points in the cluster
/// </summary>
/// <param name="clusterIndex">The current clusterId</param>
/// <returns>A new centroid for the given cluster</returns>
private KmeansVector CalculateCentroidPerCluster(int clusterIndex)
{
KmeansVector newCentroid = new KmeansVector(new double[_vectorLength]);
//Total amount of vectors in a cluster
int totalAmountVectors = 0;
//Sum all the vectors with the same cluster ID
foreach (KmeansVector vector in _vectors)
{
if (vector.ClusterId != clusterIndex)
{
continue;
}
//Add every dimension of the selected vector to the existing vector
for (int dimension = 0; dimension < vector.Coordinates.Length; dimension++)
{
newCentroid.Coordinates[dimension] = newCentroid.Coordinates[dimension] + vector.Coordinates[dimension];
}
totalAmountVectors++;
}
for (int s = 0; s < newCentroid.Coordinates.Length; s++)
{
newCentroid.Coordinates[s] = newCentroid.Coordinates[s] / totalAmountVectors;
}
return(newCentroid);
}
/// <summary>
/// Uses the Euclidean Distance to return the nearest centroid to the given vector
/// </summary>
/// <param name="vector">The vector to calculate the euclidean distance for</param>
private void FindClosestCentroid(KmeansVector vector)
{
double shortestDistance = -1;
//Find the shortest distance from the given vector to a centroid
for (int centroidId = 0; centroidId < _centroids.Count; centroidId++)
{
double distance = 0;
//For every centroid dimension - Subtract every coordinate from the centroid
for (int dimension = 0; dimension < _centroids[centroidId].Coordinates.Length; dimension++)
{
double delta = _centroids[centroidId].Coordinates[dimension] - vector.Coordinates[dimension];
distance = distance + Math.Pow(delta, 2);
}
distance = Math.Sqrt(distance);
if (shortestDistance < 0 || distance < shortestDistance)
{
shortestDistance = distance;
vector.ShortestDistanceToCentroid = distance;
vector.ClusterId = centroidId;
}
}
}
