public ClusterizationResult <double> MakeClusterization(IList <DataItem <double> > data)
{
Dictionary <double[], IList <DataItem <double> > > clusterization = new Dictionary <double[], IList <DataItem <double> > >();
Random r = new Random(DateTime.Now.Second);
double[] min = new double[data.First().Input.Length];
double[] max = new double[data.First().Input.Length];
for (int i = 0; i < data.First().Input.Length; i++)
{
min[i] = (from d in data
select d.Input[i]).Min();
max[i] = (from d in data
select d.Input[i]).Max();
}
for (int i = 0; i < _clusterCount; i++)
{
double[] v = new double[data.First().Input.Length];
for (int j = 0; j < data.First().Input.Length; j++)
{
v[j] = min[j] + r.NextDouble() * Math.Abs(max[j] - min[j]);
}
clusterization.Add(v, new List <DataItem <double> >());
}
bool convergence = true;
double lastCost = Double.MaxValue;
int iterations = 0;
while (true)
{
convergence = true;
foreach (DataItem <double> item in data)
{
var candidates = from v in clusterization.Keys
select new
{
Dist = _metrics.Calculate(v, item.Input),
Cluster = v
};
double minDist = (from c in candidates
select c.Dist).Min();
var minCandidates = from c in candidates
where c.Dist == minDist
select c.Cluster;
double[] key = minCandidates.First();
clusterization[key].Add(item);
}
double cost = 0;
List <double[]> newMeans = new List <double[]>();
foreach (double[] key in clusterization.Keys)
{
double[] v = new double[key.Length];
if (clusterization[key].Count > 0)
{
v = _metrics.GetCentroid((from x in clusterization[key]
select x.Input).ToArray());
cost += (from d in clusterization[key]
select Math.Pow(_metrics.Calculate(key, d.Input), 2)).Sum();
}
else
{
for (int j = 0; j < data.First().Input.Length; j++)
{
v[j] = min[j] + r.NextDouble() * Math.Abs(max[j] - min[j]);
}
}
newMeans.Add(v);
}
if (lastCost <= cost)
{
break;
}
iterations++;
if (iterations == _maxIterations)
{
break;
}
lastCost = cost;
clusterization.Clear();
foreach (double[] mean in newMeans)
{
clusterization.Add(mean, new List <DataItem <double> >());
}
}
return(new ClusterizationResult <double>()
{
Centroids = new List <double[]>(clusterization.Keys),
Clusterization = clusterization,
Cost = lastCost
});
}
