static bool cluster(k_Means trainingData, bool initialStep)
{
bool stop = true;
if (initialStep)
{
//loop over training data
for (int i = 0; i < trainingData.trainingSet.Count; i++)
{
double[] centersED = new double[trainingData.clustersNumber];
//cluster
for (int j = 0; j < trainingData.clustersNumber; j++)
{
centersED[j] = ecludeanDistance(trainingData.trainingSet[i].FacialGestureFeatures,
trainingData.clusters[j].first);
}
//pick min ED
int minInd = minIndex(centersED);
//add current sample to cluster with minInd
trainingData.clusters[minInd].second.Add(trainingData.trainingSet[i]);
stop = false;
}
}
else //loop over samples at clusters
{
for (int i = 0; i < trainingData.clustersNumber; i++)
{
//cluster
for (int j = 0; j < trainingData.clusters[i].second.Count; j++)
{
double[] centersED = new double[trainingData.clustersNumber];
//cluster
for (int k = 0; k < trainingData.clustersNumber; k++)
{
centersED[k] = ecludeanDistance(trainingData.clusters[i].second[j].FacialGestureFeatures,
trainingData.clusters[k].first);
}
//pick min ED
int minInd = minIndex(centersED);
if (minInd != i)
{
//not in same cluster
//add current sample to cluster with minInd and remove it from old cluster
trainingData.clusters[minInd].second.Add(trainingData.clusters[i].second[j]);
trainingData.clusters[i].second.RemoveAt(j);
stop = false;
}
}
}
}
return(stop);
}
static double[] updateCenteroids(k_Means trainingData, int clusterIndex)
{
//each center is the mean of the cluster
double[] newClusterCenter = Enumerable.Repeat((double)0, FacialGesture.facialGestureFeaturesPerSample).ToArray <double>();
for (int i = 0; i < trainingData.clusters[clusterIndex].second.Count; i++)
{
for (int j = 0; j < FacialGesture.facialGestureFeaturesPerSample; j++)
{
newClusterCenter[j] += trainingData.clusters[clusterIndex].second[i].FacialGestureFeatures[j];
}
}
for (int j = 0; j < FacialGesture.facialGestureFeaturesPerSample; j++)
{
newClusterCenter[j] /= trainingData.clusters[clusterIndex].second.Count;
}
return(newClusterCenter);
}
//////////////////////////////////////////////////////////
public NeuralNetwork(int NetworkTotalLayersNumber, int[] LayerNueronsNumber, /*int inputSize, int outputSize,*/
int activationType, double learningRate, int learningAlgoritm, bool bias = false, int RBFLayerIndex = 0)
{
this.NetworkTotalLayersNumber = NetworkTotalLayersNumber;
//this.NetworkHiddenLayersNumber = this.NetworkHiddenLayersNumber - 2 /*input and output*/;
this.LayerNueronsNumber = LayerNueronsNumber; //include input layer
this.inputSize = LayerNueronsNumber[0 /*input*/]; //number of neurons of input layer
this.outputSize = LayerNueronsNumber[NetworkTotalLayersNumber - 1]; //number of neurons of output layer
this.activationType = activationType;
this.learningRate = learningRate;
//NeuralNetwork construction
layers = new List <NetworkLayer>(this.NetworkTotalLayersNumber - 1 /*exclude input layer 34an mfeha4 processing*/);
for (int i = 0; i < this.NetworkTotalLayersNumber - 2 /*for output layer*/; i++)
{
//add hidden layer
//layer input size = size of previous layer
//layer output size = neurons number
layers.Add(new NetworkLayer(LayerNueronsNumber[i], LayerNueronsNumber[i + 1], i, activationType,
learningRate, learningAlgoritm, bias));
}
//output Layer
layers.Add(new NetworkLayer(LayerNueronsNumber[NetworkTotalLayersNumber - 2], LayerNueronsNumber[NetworkTotalLayersNumber - 1],
NetworkTotalLayersNumber - 2 /*exclude input layer*/, activationType, learningRate, learningAlgoritm, bias, true /*output layer*/));
networkErrorSignal = new double[outputSize];
this.RBFLayerIndex = RBFLayerIndex;
this.learningAlgoritm = learningAlgoritm;
this.bias = bias;
this.KMean = new k_Means(LayerNueronsNumber[RBFLayerIndex + 1], FacialGestures.trainingFacialGestures);
}
public static List <MyPair <double[] /*centroid*/, double /*variance*/> > clusterProcess(k_Means trainingData)
{
//apply k-means clustering on trainingSet and return each cluster center and variance
if (trainingData.clustersNumber <= trainingData.trainingSet.Count)
{
bool stop = false, initialeStep = true;
while (!stop)
{
//cluster process
stop = cluster(trainingData, initialeStep);
//loop over clusters
for (int i = 0; i < trainingData.clusters.Count; i++)
{
trainingData.clusters[i].first = updateCenteroids(trainingData, i);
}
initialeStep = false;
}
}
//till now we have centroids, need to calculate variance
for (int i = 0; i < trainingData.clustersNumber; i++)
{
//calculate variance of each cluster
double variance = 0;
for (int j = 0; j < trainingData.clusters[i].second.Count; j++)
{
variance += ecludeanDistance(trainingData.clusters[i].second[j].FacialGestureFeatures, trainingData.clusters[i].first);
}
trainingData.clusteringResult.Add(new MyPair <double[], double>(trainingData.clusters[i].first, variance));
}
return(trainingData.clusteringResult); //mlk4 clusters 3ndy :'D
}