//Dado un elemento y una lista de clusters devuelve la posicion del cluster mas similar
private int ClusterProcessedElement(Element e, List <Cluster> l)
{
//Esto es asi ya que estamos tratando con dissimilitud, es decir mientras mas pequeno mas cerca estan
double minDist = double.MaxValue;
int bestCluster = 0;
for (int i = 0; i < ClustersCount; i++)
{
double dist = Proximity.CalculateProximity(e, l[i].Centroid);
if (dist < minDist)
{
minDist = dist;
bestCluster = i;
}
}
return(bestCluster);
}
private string BuildInputFile(ref int acurrent)
{
string folderpath = Utils.ExesFolderPath;
string filename = "inputgraph" + Utils.GetUniqueID + ".config";
string filepath = folderpath + Path.DirectorySeparatorChar + filename;
if (File.Exists(filepath))
{
File.Delete(filepath);
}
FileStream fs = new FileStream(filepath, FileMode.CreateNew, FileAccess.ReadWrite);
StreamWriter sw = new StreamWriter(fs);
//First Line #vertex , #edges , #code
int vertices = Set.ElementsCount;
int edges = (vertices * (vertices - 1)) / 2;
int code = 1;
string firstLine = vertices + " " + edges + " " + code;
sw.WriteLine(firstLine);
int edgesZero = 0;//Este es un numero par ya que si (i,j)=0 entonces tbn sera cero (j,i)=0
//n-lines, where n is the amount of vertices, the vertices start in 1
StringBuilder line;
for (int i = 0; i < Set.ElementsCount; i++)
{
line = new StringBuilder();
for (int j = 0; j < Set.ElementsCount; j++)
{
if (i != j)
{
int adj = j + 1;
double edgeweight = Proximity.CalculateProximity(Set[i], Set[j]);
//Codigo Chiki
string number = (edgeweight * 100000).ToString();
string[] parts = number.Split('.');
int temp = 0;
if (!int.TryParse(parts[0], out temp))
{
temp = int.MaxValue;
}
//hasta aqui
if (temp != 0)
{
line.Append(adj + " " + temp + " ");
}
else
{
edgesZero++;
}
}
}
//Write the line for the vertex i+1
sw.WriteLine(line);
}
sw.Flush();
//Actualizar la cantidad de aristas ya que pudo haber aristas con costo cero
//las cuales no pertenecen al grafo ya que segun la especificacion del fichero de entrada
//el costo de las aristas debe ser mayor estricto que cero
if (edgesZero > 0)
{
fs.Seek(0, SeekOrigin.Begin);
string newFirstLine = vertices + " " + (edges - edgesZero / 2) + " " + code;
int lengthNewFirstLine = newFirstLine.Length;
int lengthOldFirstLine = firstLine.Length;
byte[] arr = new byte[lengthOldFirstLine];
for (int i = 0; i < lengthNewFirstLine; i++)
{
arr[i] = (byte)newFirstLine[i];
}
if (lengthNewFirstLine < lengthOldFirstLine)
{
byte white = (byte)' ';
for (int i = lengthNewFirstLine; i < lengthOldFirstLine; i++)
{
arr[i] = white;
}
}
fs.Write(arr, 0, lengthOldFirstLine);
fs.Flush();
}
sw.Close();
fs.Close();
return(filename);
}
public override Structuring BuildStructuring()
{
try
{
if (Set == null)
{
throw new NullReferenceException();
}
int _current = 1;
int _max = Set.ElementsCount;
if (IContainerProgressBar != null)
{
IContainerProgressBar.ResetProgressBar(1, _max, true);
IContainerProgressBar.UpdateProgressBar(1, "Running Hierarchical agglomerative algorithm...", true);
}
if (ClustersCount > Set.ElementsCount)
{
ClustersCount = Set.ElementsCount;
}
if (ClustersCount <= 0)
{
throw new Exception("La cantidad de clusters debe ser mayor que cero");
}
if (ClustersCount == 1)
{
Dictionary <string, Cluster> dic_clus = new Dictionary <string, Cluster>();
string name = "C-0";
List <Element> temp = new List <Element>();
for (int i = 0; i < Set.ElementsCount; i++)
{
if (IContainerProgressBar != null)
{
IContainerProgressBar.UpdateProgressBar(_current++, "Running Hierarchical agglomerative algorithm...", false);
}
temp.Add(Set[i]);
}
dic_clus.Add(name, new Cluster(name)
{
Elements = temp
});
Structuring = new Partition()
{
Clusters = dic_clus, Proximity = Proximity
};
if (IContainerProgressBar != null)
{
IContainerProgressBar.FinishProgressBar();
}
return(Structuring);
}
double AlfaI = 0, AlfaJ = 0, Beta = 0, Gamma = 0;
//Al inicio cada elemento es un cluster
double[,] DMatrix = new double[Set.Elements.Count, Set.Elements.Count];
List <Cluster> clusters = new List <Cluster>();
bool[] des = new bool[Set.Elements.Count];//true si ese elemento ya no es representativo
for (int i = 0; i < Set.Elements.Count; i++)
{
List <Element> l = new List <Element>();
l.Add(Set[i]);
clusters.Add(new Cluster("C-" + i, l));
}
//Construir para cada cluster una cola con prioridad
//con las disimilitudes de el con el resto de los clusters
//y la matriz de todas las disimilitudes O(n2*log n)
List <HeapArray <Container> > lh = new List <HeapArray <Container> >();
for (int i = 0; i < Set.Elements.Count; i++)
{
lh.Add(new HeapArray <Container>(Set.Elements.Count - 1));//No se pone la diss de un elemento con el mismo
}
for (int i = 0; i < Set.Elements.Count; i++)
{
for (int j = i + 1; j < Set.Elements.Count; j++)
{
double temp_diss = Proximity.CalculateProximity(Set.Elements[i], Set.Elements[j]);
DMatrix[i, j] = temp_diss;
DMatrix[j, i] = temp_diss;
lh[i].Add(new Container {
Rank = temp_diss, Name = i, Cluster = j
});
lh[j].Add(new Container {
Rank = temp_diss, Name = j, Cluster = i
});
}
}
//Algoritmo O(n2*log n)
for (int i = Set.Elements.Count; i > ClustersCount; i--)
{
if (IContainerProgressBar != null)
{
IContainerProgressBar.UpdateProgressBar(_current++, "Running Hierarchical agglomerative algorithm...", false);
}
//Seleccionar los 2 clusters mas similares O(n)
double min = double.MaxValue;
int cluster_i = 0, cluster_j = 0;
int pos_cluster_i = 0, pos_cluster_j = 0;
for (int j = lh.Count - 1; j >= 0; j--)
{
if (lh[j].First.Rank < min)
{
min = lh[j].First.Rank;
cluster_i = lh[j].First.Name;
cluster_j = lh[j].First.Cluster;
pos_cluster_i = j;
}
}
for (int j = 0; j < lh.Count; j++)
{
if (lh[j].First != null && lh[j].First.Name == cluster_j)
{
pos_cluster_j = j;
break;
}
}
//Calcular posiciones para borrar y guardar
int erase_pos = 0, final_pos = 0;
erase_pos = pos_cluster_i > pos_cluster_j ? pos_cluster_i : pos_cluster_j;
final_pos = pos_cluster_i < pos_cluster_j ? pos_cluster_i : pos_cluster_j;
lh.RemoveAt(erase_pos);
//Actualizar los parametros AlfaI, AlfaJ, Beta y Gamma
double cluster_i_count = clusters[erase_pos].ElementsCount;
double cluster_j_count = clusters[final_pos].ElementsCount;
AlfaI = UpdateAlfaI(cluster_i_count, cluster_j_count);
AlfaJ = UpdateAlfaJ(cluster_i_count, cluster_j_count);
Beta = UpdateBeta(cluster_i_count, cluster_j_count);
Gamma = UpdateGamma(cluster_i_count, cluster_j_count);
//Unir los clusters
foreach (Element item in clusters[erase_pos].Elements)
{
clusters[final_pos].AddElement(item);
}
clusters.RemoveAt(erase_pos);
//Actualizar DMatrix con la disimilitud del nuevo cluster
//y el resto de los clusters O(n)
//Formula que se usa segun el algoritmo
//La posicion del cluster i en Dmatrix es cluster_i
//La posicion del cluster j en Dmatrix es cluster_j
int pos_h = -1;
int pos_i = cluster_i;
int pos_j = cluster_j;
if (erase_pos == pos_cluster_i)
{
des[cluster_i] = true;
pos_h = cluster_j;
}
else
{
des[cluster_j] = true;
pos_h = cluster_i;
}
for (int k = 0; k < DMatrix.GetLength(0); k++)
{
DMatrix[pos_h, k] = AlfaI * DMatrix[pos_i, k] + AlfaJ * DMatrix[pos_j, k] + Beta * DMatrix[pos_i, pos_j] + Gamma * Math.Abs(DMatrix[pos_i, k] - DMatrix[pos_j, k]);
DMatrix[k, pos_h] = DMatrix[pos_h, k];
}
//Actualizar array de Heaps con la disimilitud del nuevo cluster
//y el resto de los clusters O(n*log n)
lh[final_pos] = new HeapArray <Container>(Set.Elements.Count - 1);
for (int j = 0; j < lh.Count; j++)
{
Container[] lc = lh[j].ToArray;
lh[j] = new HeapArray <Container>(Set.Elements.Count - 1);
for (int k = 1; k < lc.Length; k++)
{
if (lc[k] == null)
{
break;
}
if (lc[k].Cluster != cluster_i && lc[k].Cluster != cluster_j)
{
lh[j].Add(lc[k]);
}
}
if (j != final_pos && lh[j].First != null)
{
lh[j].Add(new Container {
Rank = DMatrix[pos_h, lh[j].First.Name], Name = lh[j].First.Name, Cluster = pos_h
});
}
}
for (int j = 0; j < DMatrix.GetLength(0); j++)
{
if (pos_h != j && pos_j != j && pos_i != j && !des[j])
{
lh[final_pos].Add(new Container {
Rank = DMatrix[pos_h, j], Name = pos_h, Cluster = j
});
}
}
}
//Crear Dictionary<string,Cluster> para construir la particion
Dictionary <string, Cluster> dic_clusters = new Dictionary <string, Cluster>();
int cont = 0;
for (int i = 0; i < clusters.Count; i++)
{
if (clusters[i] != null)
{
clusters[i].Name = "C-" + cont;
dic_clusters.Add(clusters[i].Name, clusters[i]);
cont++;
}
}
Structuring = new Partition()
{
Clusters = dic_clusters, Proximity = Proximity
};
if (IContainerProgressBar != null)
{
IContainerProgressBar.FinishProgressBar();
}
return(Structuring);
}
catch
{
if (IContainerProgressBar != null)
{
IContainerProgressBar.ShowError("Error occurred in Hierarchical agglomerative algorithm.");
}
return(null);
}
}