public KMeans(HashSet <double[]> dataSet, bool hasClass)
{
patternMatrix = new PatternMatrix();
if (hasClass)
{
foreach (var item in dataSet)
{
pattern = new Pattern();
pattern.Id = index;
pattern.AddAttributes(item);
pattern.RemoveAttributeAt(pattern.GetDimension() - 1); //remove o atributo classe da colecao de atributos
pattern.ClassAttribute = Convert.ToInt32(item[pattern.GetDimension()]); //guarda a classe do padrao
patternMatrix.AddPattern(pattern);
index++;
}
}
else
{
foreach (var item in dataSet)
{
pattern = new Pattern();
pattern.Id = index;
pattern.AddAttributes(item);
patternMatrix.AddPattern(pattern);
index++;
}
}
}
// 1 passo: calcular o centro de cada grupo.
private static Pattern ReturnClusterCenter(Cluster cluster)
{
double attributeSum = 0;
double attributeMean = 0;
double[] centroidAttribute = new double[cluster.GetPatterns()[0].GetDimension()];
for (int j = 0; j < cluster.GetPattern(0).GetDimension(); j++)
{
// -- percorre cada padrao do grupo
foreach (Pattern pattern in cluster.GetPatterns())
{
attributeSum += pattern.GetAttribute(j);
}
attributeMean = attributeSum / cluster.QuantityOfPatterns();
centroidAttribute[j] = attributeMean;
attributeSum = 0;
}
Pattern newCentroid = new Pattern();
newCentroid.AddAttributes(centroidAttribute);
return(newCentroid);
}
private int index = 0; // identificador do padrao
public KMeans(HashSet <double[]> dataSet)
{
patternMatrix = new PatternMatrix();
foreach (var item in dataSet)
{
pattern = new Pattern();
pattern.Id = index;
pattern.AddAttributes(item);
patternMatrix.AddPattern(pattern);
index++;
}
}
private void RecomputeCentroids(Cluster[] clusters, out bool changed)
{
double attributeSum = 0;
double attributeMean = 0;
Pattern newCentroid;
double[] centroidAttribute = new double[clusters[0].Centroid.GetDimension()];
changed = false;
// -- Percorre os grupos
for (int i = 0; i < clusters.Count(); i++)
{
// -- percorre cada atributo do padrao
for (int j = 0; j < clusters[0].GetPattern(0).GetDimension(); j++)
{
// -- percorre cada padrao do grupo
foreach (Pattern pattern in clusters[i].GetPatterns())
{
attributeSum += pattern.GetAttribute(j);
}
attributeMean = attributeSum / clusters[i].QuantityOfPatterns();
centroidAttribute[j] = attributeMean;
attributeSum = 0;
}
newCentroid = new Pattern();
newCentroid.AddAttributes(centroidAttribute);
newCentroid.Id = i;
// verifica se o centroide do grupo mudou. Se sim, entao atribui o novo centroide ao grupo.
if (!clusters[i].Centroid.GetAttributes().SequenceEqual(newCentroid.GetAttributes()))
{
changed = true;
clusters[i].Centroid = newCentroid; //atribui novo centroide ao grupo.
}
}
}
public PatternMatrix Transform(HashSet <double[]> dataSet)
{
Pattern patternTemp = new Pattern(); // padrao
PatternMatrix patternMatrix; // matriz de padroes
int index = 0; // identificador do padrao
PatternMatrix zPatternMatrix; //matriz de padroes normalizados
patternMatrix = new PatternMatrix();
zPatternMatrix = new PatternMatrix();
foreach (var item in dataSet)
{
patternTemp = new Pattern();
patternTemp.Id = index;
patternTemp.AddAttributes(item);
patternMatrix.AddPattern(patternTemp);
index++;
}
index = 0;
double attributeSum = 0;
double attributeMean = 0;
double variance = 0;
double attributeStandardDeviation = 0;
double zAttribute = 0;
double[] zpatternAttributes = new double[patternTemp.GetDimension()];
Pattern zPattern; //armazena o valor normalizado do padrao
foreach (Pattern pattern in patternMatrix)
{
for (int i = 0; i < pattern.GetDimension(); i++)
{
foreach (Pattern patternx in patternMatrix)
{
attributeSum += patternx.GetAttribute(i);
}
// calcula a media do atributo
attributeMean = attributeSum / patternMatrix.Size();
attributeSum = 0;
foreach (Pattern patterny in patternMatrix)
{
variance += Math.Pow((patterny.GetAttribute(i) - attributeMean), 2);
}
// calcula a variancia do atributo
variance = variance / patternMatrix.Size();
attributeStandardDeviation = Math.Sqrt(variance); // o desvio padrao eh a raiz quadrada da variancia
zAttribute = (pattern.GetAttribute(i) - attributeMean) / attributeStandardDeviation; //atributo normalizado
zpatternAttributes[i] = Math.Round(zAttribute, 2);
attributeSum = 0;
attributeMean = 0;
variance = 0;
attributeStandardDeviation = 0;
}
zPattern = new Pattern();
zPattern.AddAttributes(zpatternAttributes);
zPattern.Id = index;
zPatternMatrix.AddPattern(zPattern);
index++;
}
return(zPatternMatrix);
}
public PatternMatrix Transform(HashSet <double[]> dataSet)
{
Pattern patternTemp = new Pattern(); // padrao
PatternMatrix patternMatrix; // matriz de padroes
int index = 0; // identificador do padrao
PatternMatrix MinMaxPatternMatrix; //matriz de padroes normalizados
double minValue = Double.MaxValue;
double maxValue = Double.MinValue;
double minMaxAttribute = 0;
patternMatrix = new PatternMatrix();
MinMaxPatternMatrix = new PatternMatrix();
Pattern minMaxPattern; //armazena o padrao normalizado
foreach (var item in dataSet)
{
patternTemp = new Pattern();
patternTemp.Id = index;
patternTemp.AddAttributes(item);
patternMatrix.AddPattern(patternTemp);
index++;
}
double[] minMaxAttributes = new double[patternTemp.GetDimension()];
foreach (Pattern pattern in patternMatrix)
{
for (int i = 0; i < pattern.GetDimension(); i++)
{
foreach (Pattern patternx in patternMatrix)
{
if (patternx.GetAttribute(i) < minValue)
{
minValue = patternx.GetAttribute(i);
}
if (patternx.GetAttribute(i) > maxValue)
{
maxValue = patternx.GetAttribute(i);
}
}
minMaxAttribute = (pattern.GetAttribute(i) - minValue) / (maxValue - minValue);
minValue = Double.MaxValue;
maxValue = Double.MinValue;
minMaxAttributes[i] = Math.Round(minMaxAttribute, 2);
}
minMaxPattern = new Pattern();
minMaxPattern.AddAttributes(minMaxAttributes);
minMaxPattern.Id = index;
MinMaxPatternMatrix.AddPattern(minMaxPattern);
index++;
}
return(MinMaxPatternMatrix);
}
