static void Main(string[] args)
{
bool Kmeans = true;
bool GasSwitch = false;
bool KohonenSwitch = false;
Thread.CurrentThread.CurrentCulture = new CultureInfo("en-US");
NeuronGenerator NeuralGenerator = new NeuronGenerator();
Configuration Configuration = new Configuration();
Pointparser Parser = new Pointparser();
Parser.Filepath = "zestaw1.txt";
double InitialXRange = 2.0;
double InitialYRange = 2.0; //describes initial square (centered at 0,0) from which the neurons are generated
List <Neuron> Neurons = NeuralGenerator.GetNeurons(Configuration.GetAmountOfNeurons(), InitialXRange, InitialYRange);
List <Point> TrainingPointsList = Parser.Parse();
KohonenLearning NeuralNetworkKohonenStyle = new KohonenLearning(Neurons, TrainingPointsList, Configuration);
NeuralGas NeuralNetworkGasStyle = new NeuralGas(Neurons, TrainingPointsList, Configuration, NeuralGenerator);
VoronoiDiagramContext DiagramContext = new VoronoiDiagramContext
{
Width = 1366,
Height = 768,
MinX = TrainingPointsList.Min(point => point.XCoordinate) - 0.5,
MaxX = TrainingPointsList.Max(point => point.XCoordinate) + 0.5,
MinY = TrainingPointsList.Min(point => point.YCoordinate) - 0.5,
MaxY = TrainingPointsList.Max(point => point.YCoordinate) + 0.5,
};
if (KohonenSwitch)
{
for (int i = 0; i < 100; i++)
{
if (i % 10 == 0)
{
Console.WriteLine("Progress o 10 %");
}
NeuralNetworkKohonenStyle.Train(i);
IEnumerable <Point> Points = NeuralNetworkKohonenStyle.ReturnNeuronsAsPoints();
VoronoiDiagram.CreateImage(DiagramContext, Points, TrainingPointsList).Save("Kohonen" + i.ToString("D3") + ".png");
}
}
else if (GasSwitch)
{
for (int i = 0; i < 500; i++)
{
if (i % 50 == 0)
{
Console.WriteLine("Progress o 10 %");
}
NeuralNetworkGasStyle.Train(i);
IEnumerable <Point> Points = NeuralNetworkGasStyle.ReturnNeuronsAsPoints();
VoronoiDiagram.CreateImage(DiagramContext, Points, TrainingPointsList).Save("NeuralGas" + i.ToString("D3") + ".png");
}
}
else if (Kmeans)
{
KSrednie KMeans = new KSrednie(10, 10, 10, TrainingPointsList, 150);
Parser.ParseOut(KMeans.Clusterize(DiagramContext));
}
}
public List <Point> Clusterize(VoronoiDiagramContext Context)
{
Neurons = NeurGen.GetNeurons(ClusterCount, Xrange, Yrange);
foreach (Point Point in Points)
{
double ClosestDistance = double.MaxValue;
int BestSoFar = -1;
for (int i = 0; i < ClusterCount; i++)
{
if (Neurons[i].CalculateDistanceFrom(Point) < ClosestDistance)
{
// Console.WriteLine("Distance from Neuron" + i.ToString() + Neurons[i].CalculateDistanceFrom(Point) );
BestSoFar = i;
ClosestDistance = Neurons[i].CalculateDistanceFrom(Point);
}
}
Clusters[BestSoFar].Add(Point);
//Console.WriteLine("Assigned to: " + BestSoFar.ToString());
}
;
for (int Iteration = 0; Iteration < MaxIters; Iteration++)
{
Point Err = new Point(Iteration, 0.0);
for (int i = 0; i < ClusterCount; i++)
{
{
Double ErrorComponent = 0.0;
Point Mean = CalculateMean(Clusters[i]);
ErrorComponent = Neurons[i].CalculateDistanceFrom(Mean);
Err.YCoordinate += ErrorComponent;
//if(Iteration == 0 || Iteration == 1) Console.WriteLine(i.ToString() + Neurons[i].GetAsPoint().XCoordinate + " " + Neurons[i].GetAsPoint().YCoordinate + " MEAN " + Mean.XCoordinate + " " + Mean.YCoordinate);
Neurons[i].UpdateWeight(Mean.XCoordinate, Mean.YCoordinate);
}
Error.Add(Err);
}
;
//reassign clusters;
for (int i = 0; i < ClusterCount; i++)
{
Clusters[i] = new List <Point>();
}
foreach (Point Point in Points)
{
double ClosestDistance = double.MaxValue;
int BestSoFar = -1;
for (int i = 0; i < ClusterCount; i++)
{
if (Neurons[i].CalculateDistanceFrom(Point) < ClosestDistance)
{
BestSoFar = i;
ClosestDistance = Neurons[i].CalculateDistanceFrom(Point);
}
}
Clusters[BestSoFar].Add(Point);
}
;
List <Point> Voro = new List <Point>();
foreach (Neuron Neuron in Neurons)
{
Voro.Add(Neuron.GetAsPoint());
}
VoronoiDiagram.CreateImage(Context, Voro).Save("KMEANS" + Iteration.ToString() + "Means" + ClusterCount.ToString() + ".png");
}
return(Error);
}
public static Image CreateImage(VoronoiDiagramContext context, params IEnumerable <Point>[] inputs)
{
IEnumerable <IEnumerable <Vector> > dataPoints = inputs.Select(
input => input.Select(
point => new Vector(point.XCoordinate, point.YCoordinate)
)
);
double factorX = context.Width / (context.MaxX - context.MinX);
double factorY = context.Height / (context.MaxY - context.MinY);
double factor = Math.Min(factorX, factorY);
double minX = context.MinX - context.Width * (0.5 / factor - 0.5 / factorX);
double minY = context.MinY - context.Height * (0.5 / factor - 0.5 / factorY);
dataPoints = dataPoints.Select(points => points.Select(vector => new Vector(
(vector[0] - minX) * factor,
(vector[1] - minY) * factor
)));
PointF toPoint(Vector vector) => new PointF(
(float)vector[0],
(float)vector[1]
);
RectangleF toRectangle(Vector vector, float radius) => new RectangleF(
(float)(vector[0]) - radius,
(float)(vector[1]) - radius,
radius * 2f,
radius * 2f
);
var image = new Bitmap(context.Width, context.Height);
var linePen = new Pen(context.LineColor);
var dataPointBrushes = context.DataPointsColors.Select(color => new SolidBrush(color));
var surface = Graphics.FromImage(image);
surface.Clear(context.BackgroundColor);
surface.SmoothingMode = SmoothingMode.HighQuality;
try
{
VoronoiGraph graph = Fortune.ComputeVoronoiGraph(dataPoints.First().Distinct());
double infiniteLength = context.Width + context.Height;
foreach (var edge in graph.Edges)
{
Vector left = edge.VVertexA;
Vector right = edge.VVertexB;
if (edge.IsPartlyInfinite)
{
Vector extension = edge.DirectionVector * infiniteLength;
if (left == Fortune.VVInfinite)
{
left = edge.FixedPoint - extension;
}
if (right == Fortune.VVInfinite)
{
right = edge.FixedPoint + extension;
}
}
surface.DrawLine(linePen, toPoint(left), toPoint(right));
}
}
catch (Exception) {}
var target = dataPoints.Zip(
dataPointBrushes.Zip(
context.DataPointsRadii,
(brush, radius) => new { brush, radius }),
(points, style) => new { points, style }
).Reverse();
foreach (var data in target)
{
foreach (var point in data.points)
{
surface.FillEllipse(data.style.brush, toRectangle(point, data.style.radius));
}
}
return(image);
}