public int[] Generate(Matrix x, int k, IDistance metric)
{
if (metric == null)
metric = new EuclidianDistance();
X = x;
var means = InitializeRandom(X, k);
int[] assignments = new int[X.Rows];
for (int i = 0; i < 100; i++)
{
// Assignment step
Parallel.For(0, X.Rows, j =>
{
var min_index = -1;
var min = double.MaxValue;
for (int m = 0; m < means.Rows; m++)
{
var d = metric.Compute(X[j], means[m]);
if (d < min)
{
min = d;
min_index = m;
}
}
assignments[j] = min_index;
});
// Update Step
Matrix new_means = Matrix.Zeros(k, X.Cols);
Vector sum = Vector.Zeros(k);
// Part 1: Sum up assignments
for (int j = 0; j < X.Rows; j++)
{
int a = assignments[j];
new_means[a] += X[j, VectorType.Row];
sum[a]++;
}
// Part 2: Divide by counts
for (int j = 0; j < new_means.Rows; j++)
new_means[j] /= sum[j];
// Part 3: Check for convergence
// find norm of the difference
if ((means - new_means).Norm() < .00001)
break;
means = new_means;
}
Centers = means;
return assignments;
}
/// <summary>Generates.</summary>
/// <param name="x">The Matrix to process.</param>
/// <param name="k">The int to process.</param>
/// <param name="metric">the metric.</param>
/// <returns>An int[].</returns>
public int[] Generate(Matrix x, int k, IDistance metric)
{
if (metric == null)
{
metric = new EuclidianDistance();
}
this.X = x;
var means = this.InitializeRandom(this.X, k);
var assignments = new int[this.X.Rows];
for (var i = 0; i < 100; i++)
{
// Assignment step
Parallel.For(
0,
this.X.Rows,
j =>
{
var min_index = -1;
var min = double.MaxValue;
for (var m = 0; m < means.Rows; m++)
{
var d = metric.Compute(this.X[j], means[m]);
if (d < min)
{
min = d;
min_index = m;
}
}
// bounds?
if (min_index == -1)
{
min_index = 0;
}
assignments[j] = min_index;
});
// Update Step
var new_means = Matrix.Zeros(k, this.X.Cols);
var sum = Vector.Zeros(k);
// Part 1: Sum up assignments
for (var j = 0; j < this.X.Rows; j++)
{
var a = assignments[j];
new_means[a] += this.X[j, VectorType.Row];
sum[a]++;
}
// Part 2: Divide by counts
for (var j = 0; j < new_means.Rows; j++)
{
new_means[j] /= sum[j];
}
// Part 3: Check for convergence
// find norm of the difference
if ((means - new_means).Norm() < .00001)
{
break;
}
means = new_means;
}
this.Centers = means;
return assignments;
}
