public static double DaviesBouldinIndex(Partition p)
{
//Insure all centroids are calculated
KPoint[][] clusterPoints = new KPoint[p.Clusters.Count][];
for (int c = 0; c < p.Clusters.Count; c++)
{
clusterPoints[c] = p.GetClusterKPoints(c);
}
KPoint[] centers = p.Clusters.Select(c => new KPoint(clusterPoints[c.ClusterId])).ToArray();
double db = 0.0;
for (int i = 0; i < p.Clusters.Count; i++)
{
double Oi = averageDistanceFromCentroid(clusterPoints[i]);
double maxStat = 0.0;
for (int j = 0; j < p.Clusters.Count; j++)
{
if (i != j)
{
double Oj = averageDistanceFromCentroid(clusterPoints[j]);
double dbi = (Oi + Oj) / centers[i].elucideanDistance(centers[j]);
if (dbi > maxStat)
{
maxStat = dbi;
}
}
}
db += maxStat;
}
return(db / (double)p.Clusters.Count);
}
public int[] GetCounts(KPoint key, int linePtr)
{
if (_hiRez)
{
if (_countsRepoHiRez.ReadParts(key, _workResultHiRez))
{
return(GetOneLineFromCountsBlock(_workResultHiRez.Counts, linePtr));
}
else
{
return(null);
}
}
else
{
if (_countsRepo.ReadParts(key, _workResult))
{
return(GetOneLineFromCountsBlock(_workResult.Counts, linePtr));
}
else
{
return(null);
}
}
}
//This updates the weights of the neuron to pull towards
public void updateWeights(KPoint target, double learningRate, double influence)
{
for (int i = 0; i < weights.Dimensions; i++)
{
weights[i] += (target[i] - weights[i]) * learningRate * influence;
}
}
public void WriteWorkResult(KPoint key, MapSectionWorkResult val, bool overwriteResults)
{
// When writing include the Area's offset.
KPoint transKey = key.ToGlobal(_position);
try
{
lock (_repoLock)
{
if (overwriteResults)
{
_countsRepo.Change(transKey, val);
//WorkResultReWriteCount++;
}
else
{
_countsRepo.Add(transKey, val, saveOnWrite: false);
//WorkResultWriteCount++;
}
}
}
catch
{
Debug.WriteLine($"Could not write data for x: {transKey.X} and y: {transKey.Y}.");
}
}
public static double SilhouetteIndex(ClusteredItem datum, Partition clusters)
{
//Calculate A
double distA = 0.0;
KPoint datumPoint = clusters.Points[datum.Id];
foreach (ClusteredItem d in clusters.Clusters[datum.ClusterId].Points)
{
distA += datumPoint.elucideanDistance(clusters.Points[d.Id]);
}
distA /= (double)(clusters.Clusters[datum.ClusterId].Points.Count - 1);
double minB = double.MaxValue;
for (int i = 0; i < clusters.Clusters.Count; i++)
{
if (i != datum.ClusterId)
{
double distB = 0.0;
foreach (ClusteredItem d in clusters.Clusters[i].Points)
{
distB += datumPoint.elucideanDistance(clusters.Points[d.Id]);
}
distB /= (double)(clusters.Clusters[i].Points.Count);
if (distB < minB)
{
minB = distB;
}
}
}
return((minB - distA) / Math.Max(minB, distA));
}
public SelfOrganizingMap(PointSet data, List <int> dimensions, int totalItt, double learningRate)
{
//Get initial learning rate
this.initialLearningRate = learningRate;
this.learningRate = learningRate;
influence = 0;
this.totalItterations = totalItt;
itterations = 1;
this.dimensions = dimensions;
this.data = data;
rng = new Random((int)DateTime.Now.Ticks);
int numNeurons = 1;
foreach (int d in dimensions)
{
numNeurons *= d;
}
int[] pos = new int[dimensions.Count];
for (int i = 0; i < dimensions.Count; i++)
{
pos[i] = 0;
}
//Scale each attribute from [0,1] and store the conversion matrix
conversionMatrix = data.GetMinMaxWeights().Max.Coordinates;
foreach (KPoint d in data.PointList)
{
d.Normalize(conversionMatrix);
}
//Here We will initialize Our Neurons
neurons = new List <SOMNeuron>();
KPoint zero = KPoint.Zero(data[0].Dimensions);
KPoint one = KPoint.One(data[0].Dimensions);
for (int i = 0; i < numNeurons; i++)
{
//Generate our neurons
double[] posNeuron = new double[dimensions.Count];
for (int d = 0; d < dimensions.Count; d++)
{
posNeuron[d] = (double)pos[d];
}
SOMNeuron neuron = new SOMNeuron(zero, one, rng, new KPoint(posNeuron), i);
neurons.Add(neuron);
addOneCarry(ref pos, dimensions);
}
//Get the max radius
setRadius();
this.timeConstant = (double)totalItterations / Math.Log(radius);
}
public BallTreeNode(PointSet Points)
{
//if there is only a single point, set this as the data
if (Points.Count == 1)
{
Point = Points[0];
}
else
{
//Find the dimension of greatest spread
var minMax = Points.GetMinMaxWeights();
double maxSpread = double.MinValue;
int maxDim = 0;
for (int d = 0; d < Points.Dimensions; d++)
{
double range = minMax.Max[d] - minMax.Min[d];
if (range > maxSpread)
{
maxSpread = d;
maxSpread = range;
}
}
PivotDim = maxDim;
//Pivot Select, Replace with QuickSelect
double[] pivots = new double[Points.Count];
for (int i = 0; i < Points.Count; i++)
{
pivots[i] = Points[i][PivotDim];
}
Array.Sort(pivots);
double pivotValue = pivots[(Points.Count - 1) / 2];
//Create left and right children
List <KPoint> left = new List <KPoint>();
List <KPoint> right = new List <KPoint>();
for (int i = 0; i < Points.Count; i++)
{
if (Points[i][PivotDim] <= pivotValue)
{
left.Add(Points[i]);
}
else
{
right.Add(Points[i]);
}
}
leftNode = new BallTreeNode(new PointSet(left));
rightNode = new BallTreeNode(new PointSet(right));
}
}
public bool RetrieveWorkResultFromRepo(KPoint key, MapSectionWorkResult workResult)
{
// When writing include the Area's offset.
KPoint transKey = key.ToGlobal(_position);
lock (_repoLock)
{
bool result = _countsRepo.ReadParts(transKey, workResult);
return(result);
}
}
public bool ContainsKey(KPoint key)
{
if (_hiRez)
{
return(_countsRepoHiRez.ContainsKey(key));
}
else
{
return(_countsRepo.ContainsKey(key));
}
}
private void MainCanvas_Paint(object sender, PaintEventArgs e)
{
var x0 = MainCanvas.Width / 2;
var y0 = MainCanvas.Height / 2;
e.Graphics.DrawLine(Pens.Black, new Point(x0, 0), new Point(x0, MainCanvas.Height));
e.Graphics.DrawLine(Pens.Black, new Point(0, y0), new Point(MainCanvas.Width, y0));
center = new KPoint {
x = x0, y = y0
};
}
private CanvasSize GetImageSizeInBlocks(CountsRepoReader countsRepo)
{
bool foundMax = false;
int w = 10;
int h = 0;
KPoint key = new KPoint(w, h);
foundMax = !countsRepo.ContainsKey(key);
if (foundMax)
{
return(new CanvasSize(0, 0));
}
// Find max value where w and h are equal.
while (!foundMax)
{
w++;
h++;
key = new KPoint(w, h);
foundMax = !countsRepo.ContainsKey(key);
}
w--;
h--;
foundMax = false;
// Find max value of h
while (!foundMax)
{
h++;
key = new KPoint(w, h);
foundMax = !countsRepo.ContainsKey(key);
}
h--;
foundMax = false;
// Find max value of h
while (!foundMax)
{
w++;
key = new KPoint(w, h);
foundMax = !countsRepo.ContainsKey(key);
}
//w--;
return(new CanvasSize(w, ++h));
}
private MapSectionWorkResult RetrieveWorkResultFromRepo(KPoint key, Job localJob, bool readZValues)
{
MapSectionWorkResult workResult = GetEmptyResult(readZValues, Job.SECTION_WIDTH, Job.SECTION_HEIGHT);
if (localJob.RetrieveWorkResultFromRepo(key, workResult))
{
return(workResult);
}
else
{
return(null);
}
}
private MapSectionWorkResult GetEmptyResult(KPoint key)
{
//if(area.Size.W != SECTION_WIDTH || area.Size.H != SECTION_HEIGHT)
//{
// Debug.WriteLine("Wrong Area.");
//}
if (_emptyResult == null)
{
_emptyResult = new MapSectionWorkResult(SECTION_WIDTH * SECTION_HEIGHT, hiRez: false, includeZValuesOnRead: false);
}
return(_emptyResult);
}
private MapSectionResult ProcessSubJob(SubJob subJob)
{
if (subJob.ParentJob.CancelRequested)
{
Debug.WriteLine("Not Processing Sub Job.");
return(null);
}
if (!(subJob.ParentJob is Job localJob))
{
throw new InvalidOperationException("When processing a subjob, the parent job must be implemented by the Job class.");
}
MapSectionWorkRequest mswr = subJob.MapSectionWorkRequest;
KPoint key = new KPoint(mswr.HPtr, mswr.VPtr);
MapSectionWorkResult workResult = RetrieveWorkResultFromRepo(key, localJob, readZValues: false);
MapSectionResult result;
if (workResult == null)
{
result = CalculateMapValues(subJob, localJob, ref workResult);
}
else
{
if (workResult.IterationCount == 0 || workResult.IterationCount == mswr.MaxIterations)
{
// The WorkResult read from file has the correct iteration count. (Or we are not tracking the interation count.)
result = new MapSectionResult(localJob.JobId, mswr.MapSection, workResult.Counts);
}
else if (workResult.IterationCount < mswr.MaxIterations)
{
// Fetch the entire WorkResult with ZValues
workResult = RetrieveWorkResultFromRepo(key, localJob, readZValues: true);
// Use the current work results to continue calculations to create
// a result with the target iteration count.
result = CalculateMapValues(subJob, localJob, ref workResult);
}
else
{
throw new InvalidOperationException("Cannot reduce the number of iterations of an existing job.");
}
}
return(result);
}
//public int WorkResultWriteCount = 0;
//public int WorkResultReWriteCount = 0;
public Job(SMapWorkRequest sMapWorkRequest) : base(sMapWorkRequest)
{
_position = new KPoint(sMapWorkRequest.Area.SectionAnchor.X, sMapWorkRequest.Area.SectionAnchor.Y);
SamplePoints = GetSamplePoints(sMapWorkRequest);
Reset();
string filename = RepoFilename;
Debug.WriteLine($"Creating new Repo. Name: {filename}, JobId: {JobId}.");
_countsRepo = new ValueRecords <KPoint, MapSectionWorkResult>(filename, useHiRezFolder: false);
//Debug.WriteLine($"Starting to get histogram for {RepoFilename} at {DateTime.Now.ToString(DiagTimeFormat)}.");
//Dictionary<int, int> h = GetHistogram();
//Debug.WriteLine($"Histogram complete for {RepoFilename} at {DateTime.Now.ToString(DiagTimeFormat)}.");
}
private SOMNeuron GetBestMatchingUnit(KPoint data)
{
SOMNeuron best = null;
double dist = Double.MaxValue;
foreach (SOMNeuron n in neurons)
{
double d = n.weights.distanceSquared(data);
if (d < dist)
{
dist = d;
best = n;
}
}
return best;
}
public KMeans(PointSet points, int numClusters)
{
//Assign the points and number of clusters
Points = points;
NumClusters = numClusters;
//Create our dataPoints
ClusteredPoints = new List <ClusteredItem>();
for (int i = 0; i < points.Count; i++)
{
ClusteredPoints.Add(new ClusteredItem(i));
}
//Create a new RNG
Random rng = new Random();
//Now we find the min and max
var minMax = points.GetMinMaxWeights();
//Now we generate our clusters
//Make random clusters until we have 3 valid clusters
_clusters = new Cluster[numClusters];
_clusterLocation = new KPoint[numClusters];
do
{
for (int c = 0; c < numClusters; c++)
{
_clusters[c] = new Cluster(c);
_clusterLocation[c] = new KPoint(minMax.Min, minMax.Max, rng);
}
AssignPoints();
} while (!ValidClusters());
//KMeans Algorithm algo starts
//1. assign points
//2. calculate new centers
do
{
foreach (Cluster cl in _clusters)
{
if (cl.Points.Count > 0)
{
_clusterLocation[cl.ClusterId] = new KPoint(cl.Points.Select(i => points[i.Id]).ToArray());
}
}
NumItterations++;
} while (AssignPoints());
}
public MinHeapPriorityQueue <KPoint> NearestNeighbors(KPoint p, int k, MinHeapPriorityQueue <KPoint> Q, BallTreeNode B)
{
if (p[B.PivotDim] > Q.peek().elucideanDistance(p))
{
return(Q);
}
else if (B.IsLeaf())
{
if (p.elucideanDistance(B.Point) < p.elucideanDistance(Q.peek()))
{
if (Q.Count >= k)
{
Q.extractMin();
}
}
}
return(Q);
}
private MapSectionResult CalculateMapValues(SubJob subJob, Job localJob, ref MapSectionWorkResult workResult)
{
MapSectionWorkRequest mswr = subJob.MapSectionWorkRequest;
bool overwriteResults;
if (workResult == null)
{
workResult = BuildInitialWorkingValues(mswr);
overwriteResults = false;
}
else
{
overwriteResults = true;
}
double[] xValues = localJob.SamplePoints.XValueSections[mswr.HPtr];
double[] yValues = localJob.SamplePoints.YValueSections[mswr.VPtr];
//DateTime t0 = DateTime.Now;
//workResult = _mapCalculator.GetWorkingValues(xValues, yValues, mswr.MaxIterations, workResult);
//DateTime t1 = DateTime.Now;
//int[] testCounts = new int[10000];
//DateTime t2 = DateTime.Now;
//_mapCalculator.ComputeApprox(xValues, yValues, mswr.MaxIterations, testCounts);
//List<int> misMatcheIndexes = GetMismatchCount(workResult.Counts, testCounts, out double avgGap);
//DateTime t3 = DateTime.Now;
//Debug.WriteLine($"Block: v={mswr.VPtr}, h={mswr.HPtr} has {misMatcheIndexes.Count} mis matches, avgGap={avgGap} and avg {GetAverageCntValue(workResult.Counts)}.");
//Debug.WriteLine($"Standard: {GetElaspedTime(t0, t1)}, Approx: {GetElaspedTime(t2, t3)}.");
//_mapCalculator.ComputeApprox(xValues, yValues, mswr.MaxIterations, workResult.Counts);
workResult = _mapCalculator.GetWorkingValues(xValues, yValues, mswr.MaxIterations, workResult);
KPoint key = new KPoint(mswr.HPtr, mswr.VPtr);
localJob.WriteWorkResult(key, workResult, overwriteResults);
MapSectionResult msr = new MapSectionResult(localJob.JobId, mswr.MapSection, workResult.Counts);
return(msr);
}
public void Epoch(KPoint p)
{
//get a random item
SOMNeuron bmu = GetBestMatchingUnit(p);
neighborhoodRadius = radius * Math.Exp(-(double)itterations / timeConstant);
double neighborSqr = neighborhoodRadius * neighborhoodRadius;
QuadTreePointStruct center = new QuadTreePointStruct()
{
Index = -1,
X = bmu.position[0],
Y = bmu.position[1]
};
QuadTreePointStruct halfLength = new QuadTreePointStruct()
{
Index = -1,
X = neighborhoodRadius,
Y = neighborhoodRadius
};
List <int> neuronsInArea =
neuronQuadTree.QueryRange(new QuadTreeBoundingBox()
{
Center = center, HalfLength = halfLength
});
//Now we need to update each neuron
foreach (int i in neuronsInArea)
{
SOMNeuron n = neurons[i];
double distToBMU = n.position.elucideanDistance(bmu.position);
if (distToBMU <= neighborhoodRadius)
{
influence = Math.Exp(-(distToBMU * distToBMU) / (2 * neighborSqr));
n.updateWeights(p, learningRate, influence);
}
}
learningRate = this.initialLearningRate * Math.Exp(-(double)itterations / (double)this.totalItterations);
itterations++;
}
public void Epoch(KPoint p)
{
//get a random item
SOMNeuron bmu = GetBestMatchingUnit(p);
neighborhoodRadius = radius * Math.Exp(-(double)itterations / timeConstant);
double neighborSqr = neighborhoodRadius * neighborhoodRadius;
//Now we need to update each neuron
foreach (SOMNeuron n in neurons)
{
double distToBMUSqr = n.position.elucideanDistance(bmu.position);
if (distToBMUSqr <= neighborSqr)
{
influence = Math.Exp(-(distToBMUSqr) / (2 * neighborSqr));
n.updateWeights(p, learningRate, influence);
}
}
learningRate = this.initialLearningRate * Math.Exp(-(double)itterations / (double)this.totalItterations);
itterations++;
}
public static double getSqrdErrorDistortion(Partition p)
{
//Insure all centroids are calculated
KPoint[][] clusterPoints = new KPoint[p.Clusters.Count][];
for (int c = 0; c < p.Clusters.Count; c++)
{
clusterPoints[c] = p.GetClusterKPoints(c);
}
KPoint[] centers = p.Clusters.Select(c => new KPoint(clusterPoints[c.ClusterId])).ToArray();
double sum = 0.0;
for (int i = 0; i < p.Clusters.Count; i++)
{
for (int j = 0; j < clusterPoints[i].Length; j++)
{
sum += centers[i].distanceSquared(clusterPoints[i][j]);
}
}
return(sum / (double)p.GetClusteredItemCount());
}
public static double DunnIndex(Partition p, DunnInterDistanceType distType, DunnIntraDistanceType intraDistType)
{
//Insure all centroids are calculated
KPoint[][] clusterPoints = new KPoint[p.Clusters.Count][];
for (int c = 0; c < p.Clusters.Count; c++)
{
clusterPoints[c] = p.GetClusterKPoints(c);
}
KPoint[] centers = p.Clusters.Select(c => new KPoint(clusterPoints[c.ClusterId])).ToArray();
//Between 2 clusters(center)
double minInter = double.MaxValue;
for (int i = 0; i < p.Clusters.Count - 1; i++)
{
for (int j = i + 1; j < p.Clusters.Count; j++)
{
double distIJ = 0.0;
if (distType == DunnInterDistanceType.CentroidDist)
{
distIJ = centers[i].elucideanDistance(centers[j]);
}
else if (distType == DunnInterDistanceType.MaxDist)
{
distIJ = getInterClusterDist(clusterPoints[i], clusterPoints[j], true);
}
else if (distType == DunnInterDistanceType.MinDist)
{
distIJ = getInterClusterDist(clusterPoints[i], clusterPoints[j], false);
}
else if (distType == DunnInterDistanceType.AverageDist)
{
double sumDist = 0.0;
foreach (KPoint k in clusterPoints[i])
{
foreach (KPoint l in clusterPoints[j])
{
sumDist += k.elucideanDistance(l);
}
}
distIJ = sumDist / (double)(clusterPoints[i].Length * clusterPoints[j].Length);
}
else if (distType == DunnInterDistanceType.AverageToCentroid)
{
double distCentI = 0.0;
foreach (KPoint pointJ in clusterPoints[j])
{
distCentI += centers[i].elucideanDistance(pointJ);
}
distCentI /= (double)clusterPoints[j].Length;
double distCentJ = 0.0;
foreach (KPoint pointI in clusterPoints[i])
{
distCentJ += centers[j].elucideanDistance(pointI);
}
distCentJ /= (double)clusterPoints[i].Length;
distIJ = Math.Min(distCentI, distCentJ);
}
minInter = (distIJ < minInter) ? distIJ : minInter;
}
}
//Pairwise
double maxIntraClusterDistance = 0.0;
if (intraDistType == DunnIntraDistanceType.Complete)
{
foreach (Cluster c in p.Clusters)
{
for (int i = 0; i < c.Points.Count - 1; i++)
{
var pointI = clusterPoints[c.ClusterId][i];
for (int j = i + 1; j < c.Points.Count; j++)
{
var pointJ = clusterPoints[c.ClusterId][j];
double distIJ = pointI.elucideanDistance(pointJ);
maxIntraClusterDistance = (distIJ > maxIntraClusterDistance) ? distIJ : maxIntraClusterDistance;
}
}
}
}
else if (intraDistType == DunnIntraDistanceType.Average)
{
foreach (Cluster c in p.Clusters)
{
double sum = 0.0;
for (int i = 0; i < clusterPoints[c.ClusterId].Length - 1; i++)
{
for (int j = i + 1; j < clusterPoints[c.ClusterId].Length; j++)
{
sum += clusterPoints[c.ClusterId][i].elucideanDistance(clusterPoints[c.ClusterId][j]);
}
}
sum /= (double)(clusterPoints[c.ClusterId].Length * (clusterPoints[c.ClusterId].Length - 1));
maxIntraClusterDistance = (sum > maxIntraClusterDistance) ? sum : maxIntraClusterDistance;
}
}
else if (intraDistType == DunnIntraDistanceType.AverageToCentroid)
{
foreach (Cluster c in p.Clusters)
{
double sum = 0.0;
foreach (KPoint point in clusterPoints[c.ClusterId])
{
sum += 2 * centers[c.ClusterId].elucideanDistance(point);
}
sum /= (double)(clusterPoints[c.ClusterId].Length);
maxIntraClusterDistance = (sum > maxIntraClusterDistance) ? sum : maxIntraClusterDistance;
}
}
double dunn = minInter / maxIntraClusterDistance;
return(dunn);
}
//Initialize tyhe weights with a randon between min and max
public SOMNeuron(KPoint min, KPoint max, Random rng, KPoint position, int id)
{
weights = new KPoint(min, max, rng);
this.position = position;
this.id = id;
}
public BallTree(KPoint points)
{
}
private void MouseMoveEvent(object s, MouseEventArgs e) {
if (dragging && e.Button == MouseButtons.Left) {
Left = mouseDownWindowLocation.x + (e.X - mouseDownLocation.x);
Top = mouseDownWindowLocation.y + (e.Y - mouseDownLocation.y);
mouseDownWindowLocation = new KPoint(Left, Top);
}
}
public void Build(string fn, bool hiRez)
{
// TODO: HiRez, blockWidth and blockHeight should come from the RepoFile.
MapInfoWithColorMap miwcm = ReadFromJson(fn);
int maxIterations = miwcm.MapInfo.MaxIterations;
ColorMap colorMap = miwcm.ColorMap;
string repofilename = miwcm.MapInfo.Name;
//ValueRecords<KPoint, MapSectionWorkResult> countsRepo = new ValueRecords<KPoint, MapSectionWorkResult>(repofilename, useHiRezFolder: hiRez);
//int blockLength = BlockWidth * BlockHeight;
//MapSectionWorkResult workResult = new MapSectionWorkResult(blockLength, hiRez: hiRez, includeZValuesOnRead: false);
//CanvasSize imageSizeInBlocks = GetImageSizeInBlocks(countsRepo);
int blockLength = BlockWidth * BlockHeight;
CountsRepoReader countsRepoReader = new CountsRepoReader(repofilename, hiRez, BlockWidth, BlockHeight);
CanvasSize imageSizeInBlocks = GetImageSizeInBlocks(countsRepoReader);
int w = imageSizeInBlocks.Width;
int h = imageSizeInBlocks.Height;
CanvasSize imageSize = new CanvasSize(w * BlockWidth, h * BlockHeight);
string imagePath = GetImageFilename(fn, imageSize.Width, hiRez, BasePath);
KPoint key = new KPoint(0, 0);
using (PngImage pngImage = new PngImage(imagePath, imageSize.Width, imageSize.Height))
{
for (int vBPtr = 0; vBPtr < h; vBPtr++)
{
key.Y = vBPtr;
for (int lPtr = 0; lPtr < 100; lPtr++)
{
ImageLine iLine = pngImage.ImageLine;
int linePtr = vBPtr * BlockHeight + lPtr;
for (int hBPtr = 0; hBPtr < w; hBPtr++)
{
key.X = hBPtr;
//if (countsRepo.ReadParts(key, workResult))
//{
// int[] allCounts = workResult.Counts;
// int[] countsForThisLine = GetOneLineFromCountsBlock(allCounts, lPtr);
// BuildPngImageLineSegment(hBPtr * BlockWidth, countsForThisLine, iLine, maxIterations, colorMap);
//}
//else
//{
// BuildBlankPngImageLineSegment(hBPtr * BlockWidth, BlockWidth, iLine);
//}
int[] countsForThisLine = countsRepoReader.GetCounts(key, lPtr);
if (countsForThisLine != null)
{
BuildPngImageLineSegment(hBPtr * BlockWidth, countsForThisLine, iLine, maxIterations, colorMap);
}
else
{
BuildBlankPngImageLineSegment(hBPtr * BlockWidth, BlockWidth, iLine);
}
}
pngImage.WriteLine(iLine);
}
}
}
}
private void MouseDownEvent(object s, MouseEventArgs e) {
if (e.Button == MouseButtons.Left) {
mouseDownLocation = new KPoint(e.Location.X, e.Location.Y);
mouseDownWindowLocation = new KPoint(Left, Top);
dragging = true;
}
}
public static double averageDistanceFromCentroid(KPoint[] points)
{
KPoint center = new KPoint(points);
return(points.Select(p => p.elucideanDistance(center)).Sum() / (double)points.Length);
}
public HexagonalSelfOrganizingMap(PointSet data, int dimension, double learningRate)
{
_dimension = dimension;
//Get initial learning rate
this.initialLearningRate = learningRate;
this.learningRate = learningRate;
influence = 0;
itterations = 1;
this.data = data;
rng = new Random();
int numNeurons = dimension * dimension;
//Scale each attribute from [0,1] and store the conversion matrix
conversionMatrix = data.GetMinMaxWeights().Max.Coordinates;
foreach (KPoint d in data.PointList)
{
d.Normalize(conversionMatrix);
}
//Here We will initialize Our Neurons
neurons = new List <SOMNeuron>();
KPoint zero = KPoint.Zero(data[0].Dimensions);
KPoint one = KPoint.One(data[0].Dimensions);
double halfNeuronDelta = 1 / Math.Sqrt(3.0);
//Initialize our Quadtree for reference
double centerX = (dimension - 0.5) * (2 * halfNeuronDelta) / 2.0;
double centerY = (dimension - 1.0) / 2.0;
QuadTreePointStruct center = new QuadTreePointStruct()
{
Index = -1, X = centerX, Y = centerY
};
QuadTreePointStruct halfDistance = new QuadTreePointStruct()
{
Index = -1, X = centerX + halfNeuronDelta, Y = centerY + 0.5
};
neuronQuadTree = new QuadTree(new QuadTreeBoundingBox()
{
Center = center, HalfLength = halfDistance
});
int neuronIndex = 0;
for (int r = 0; r < dimension; r++)
{
double xPos = (r % 2 == 1) ? halfNeuronDelta : 0.0;
for (int c = 0; c < dimension; c++)
{
//Generate our neurons
double[] posNeuron = { xPos, (double)r };
SOMNeuron neuron = new SOMNeuron(zero, one, rng, new KPoint(posNeuron), neuronIndex);
QuadTreePointStruct neuronQTPos = new QuadTreePointStruct()
{
Index = neuronIndex,
X = posNeuron[0],
Y = posNeuron[1]
};
neuronQuadTree.Insert(neuronQTPos);
neurons.Add(neuron);
xPos += 2 * halfNeuronDelta;
neuronIndex++;
}
}
//Get the max radius
SetRadius();
timeConstant = (double)totalItterations / Math.Log(radius);
}
