SOMCode: GraphSegmentation.h Source File

00001 #ifndef GraphSegmentation_H
00002 #define GraphSegmentation_H
00003 
00004 
00005 //#include <TeAdoDB.h>
00006 //#include <TeDatabase.h>
00007 
00008 
00009 #include "Som.h"
00010 
00011 class GraphSegmentation  {
00012 public:
00013 
00014         
00015         GraphSegmentation(){};
00016 
00017         
00018         TIntVector Segment( SOM* som, int min_neurons_per_cluster );
00019 
00020         Value_Type  Davies_Bouldin( Value_Type p, Value_Type q );  
00021 
00022         Value_Type  Davies_Bouldin_Data( Value_Type p, Value_Type q );  
00023 
00024         int  getNumClusters() { return _num_clusters; };
00025 
00026         TIntVector Siponen();
00027 
00028         //void createField( vector<string> params );
00029 
00030 
00031 private:
00032 
00033         void Init( SOM* som );
00034 
00036         TMatrix Centroid( );
00037 
00038         TMatrix Distance( );
00039 
00040         TVector Winner( );
00041 
00042         TVector MeanDistance( );
00043 
00044         TMatrix AdjacentNode( );
00045 
00046         Value_Type    Hmin( );
00047 
00048         void  Label( int i, int& group, TIntVector& vec, bool& changed );
00049         
00050         
00051 private:
00052         SOM * _som;
00053         
00054         TMatrix _adjacentNode;
00055         TMatrix _distance;
00056         TVector _winner;
00057         TVector _meanDistance;
00058         TMatrix _centroid;
00059 
00060         Value_Type      _Hmin;
00061         Value_Type      _delta;
00062         int             m;
00063 
00064         TIntVector  _clusters;
00065         TVector    _S;
00066         TMatrix    _M;
00067         int                _num_clusters;
00068         TIntVector _num_vector_cluster;
00069         TMatrix    _cluster_centroid;
00070         TMatrix    _R;
00071         
00072 };
00073 
00074 /*
00075 void 
00076 GraphSegmentation::createField( vector<string> params ) {
00077 
00078         string host = params[0];
00079     string dbname = params[1];
00080     string user = params[2];
00081     string pass = params[3];
00082         string tableName = params[4];
00083         string fieldName = params[5];
00084         string labelField = params[6];
00085 
00086         TLabel label = _som->getMapcode()->getLabel();
00087         
00088 
00089         int i;
00090 
00091         // Opens a connection to a database accessible though ADO
00092     TeDatabase* db = new TeAdo();
00093     if (!db->connect(host,user,pass,dbname,0))
00094     {
00095                cout << "Error: " << db->errorMessage() << endl;
00096                return;
00097     }
00098 
00099         TeAttribute atr;
00100         atr.rep_.name_= fieldName;
00101     atr.rep_.type_ = TeREAL;
00102     atr.rep_.numChar_ = 15;
00103     atr.rep_.decimals_ = 0;
00104     
00105         if (!db->addColumn(tableName, atr.rep_))
00106         {
00107                 cout << "Error adding column to table: " << db->errorMessage() << endl;
00108                 db->close();
00109                 return;
00110         }
00111 
00112 
00113 
00114     TeDatabasePortal* portal = db->getPortal();
00115     if (!portal)
00116     {
00117             cout << "Error getting a portal to dabase: " << db->errorMessage() << endl;
00118             db->close();
00119             return;
00120     }
00121                                 
00122         string sql = "UPDATE " + tableName  + " SET " + fieldName + " = 0";
00123         
00124 
00125         if (portal->query(sql))
00126         {
00127                   cout << "Error submiting query: " << db->errorMessage() << endl;
00128           delete portal;
00129           db->close();
00130           return;
00131         }
00132         
00133 
00134         for (i=0; i< m; i++ ) {
00135                 for (int j = 0; j < label[i].size(); j++ ) {
00136                         string value = inttostr(_clusters[i+1]);
00137                         string id    = label[i][j];
00138                         sql = "UPDATE " + tableName  + " SET " + fieldName + " = " + value + " WHERE " + labelField + " = " + "\""+ id + "\"";
00139                         //cout << sql << endl;
00140                         if (portal->query(sql))
00141                         {
00142                                 cout << "Error submiting query: " << db->errorMessage() << endl;
00143                                 delete portal;
00144                                 db->close();
00145                                 return;
00146                         }
00147                         
00148                 }
00149         }
00150 
00151         delete portal;
00152     db->close();
00153         
00154 };
00155 
00156 */
00157 TIntVector 
00158 GraphSegmentation::Siponen() 
00159 {
00160         int i, j, k;
00161         int d   =  _som->getMapcode()->getNumVar();
00162         TMatrix codebook        =  _som->getMapcode()->getCodebook();
00163 
00164         TIntVector s_vector =  create_intvector( 1, _num_clusters );
00165         TMatrix max_value       = create_matrix( 1, _num_clusters, 1, d );
00166         TMatrix min_value       = create_matrix( 1, _num_clusters, 1, d );
00167         TMatrix mean_value      = create_matrix( 1, _num_clusters, 1, d );
00168         TVector count           = create_vector( 1, _num_clusters );
00169 
00170         for (i=1; i <= _num_clusters; i++) {
00171                 count[i] = 0;
00172                 s_vector[i] = 0;
00173                 for (j=1; j <= d; j++) {
00174                         max_value[i][j] = -99999;
00175                         min_value[i][j] =  99999;
00176                         mean_value[i][j] = 0;
00177                 }
00178         }
00179 
00180         for (i=1; i<= _num_clusters; i++ ) {
00181                 for (j=1; j<=m; j++ ) {
00182                         if ( _clusters[j] == i ) {
00183                                 for (k=1; k<=d; k++ ) {
00184                                         if ( codebook[j][k] > max_value[i][k] )
00185                                                 max_value[i][k] = codebook[j][k];
00186                                         if ( codebook[j][k] < min_value[i][k] )
00187                                                 min_value[i][k] = codebook[j][k];
00188                                         mean_value[i][k] += codebook[j][k]; 
00189                                 }       
00190                                 count[i]++;
00191                         }
00192                 }
00193         }
00194 
00195         for (i=1; i <= _num_clusters; i++)
00196                 for (k=1; k <= d; k++) {
00197                         if ( count[i] != 0 )
00198                                 mean_value[i][k] /= count[i];
00199                         else
00200                                 mean_value[i][k] = 0;
00201                 }
00202 
00203         TMatrix s_v = create_matrix( 1, _num_clusters, 1, d );
00204         for (i=1; i <= _num_clusters; i++)
00205                 for (k=1; k <= d; k++)  {
00206                         if ( (( max_value[i][k] - min_value[i][k] )) != 0 )
00207                                 s_v[i][k] =  ( mean_value[i][k] - min_value[i][k] )/( max_value[i][k] - min_value[i][k] );
00208                         else
00209                                 s_v[i][k] = 0;
00210                 }
00211 
00212         Value_Type sum;
00213         TMatrix s_vc = create_matrix( 1, _num_clusters, 1, d );
00214         for (i=1; i <= _num_clusters; i++)
00215                 for (k=1; k <= d; k++)  {
00216                         sum = 0;
00217                         for (j=1; j<= _num_clusters; j++ )
00218                                 if ( j != i )
00219                                         sum += s_v[j][k];
00220                         if ( sum != 0 )
00221                                 s_vc[i][k] = s_v[i][k] / ( ( 1.0/((Value_Type)_num_clusters - 1.0) )*sum );
00222                         else
00223                                 s_vc[i][k] = 0;
00224                 }
00225 
00226 
00227         //printmatrix( s_v, 1, _num_clusters, 1, d );
00228         //printmatrix( s_vc, 1, _num_clusters, 1, d );
00229 
00230         Value_Type s_max_value;
00231         
00232         for (i=1; i<= _num_clusters; i++ ) {
00233                 s_max_value = -1.0;
00234                 for (k=1; k <= d; k++)  {
00235                         if (s_vc[i][k] > s_max_value ) {
00236                                 s_vector[i] = k;
00237                                 s_max_value = s_vc[i][k];
00238                         }
00239                 }
00240         }
00241 
00242         return s_vector;
00243 };
00244 
00245 Value_Type  
00246 GraphSegmentation::Davies_Bouldin_Data( Value_Type p, Value_Type q )
00247 {
00248         int ii, i, j, k, l;
00249         int d  = _som->getMapcode()->getNumVar();
00250         TMatrix entries = _som->getData()->getEntries();
00251 
00252         TVector         _S                                      = create_vector( 1, _num_clusters );
00253         TMatrix         _cluster_centroid       = create_matrix(1, _num_clusters, 1, d );
00254         TIntVector      _num_vector_cluster = create_intvector( 1, _num_clusters );
00255         TMatrix         _M                                      = create_matrix( 1, _num_clusters,1, _num_clusters );
00256         TMatrix         _R                                      = create_matrix( 1, _num_clusters,1, _num_clusters );
00257 
00258         TvecLabel entries_label                 = _som->getData()->getId(); //getLabel();
00259         TLabel  codebook_label                  = _som->getMapcode()->getLabel();
00260         int n                                                   = _som->getData()->getDataSize();                                               
00261 
00262         for (i=1; i<= _num_clusters; i++ ) {
00263                 _S[i] = 0.0; 
00264                 _num_vector_cluster[i] = 0;
00265                 for (j=1; j <= _num_clusters; j++ )     _R[i][j] = _M[i][j] = 0.0;
00266                 for (j=1; j<= d; j++ )  _cluster_centroid[i][j] = 0.0;
00267         }
00268 
00269         // Calculo dos centroides <<Dados de entrada>>
00270         TVector count = create_vector( 1, _num_clusters ); for (i=1;i<=_num_clusters; i++) count[i] = 0.0;
00271         for (i=1; i<=m; i++ ) {
00272                 if ( _clusters[i] > 0 ) {
00273                         for (l=0; l< codebook_label[i-1].size(); l++ )
00274                                 for (j=1; j<=n; j++) {
00275                                         if ( codebook_label[i-1][l] == entries_label[j-1] ) {
00276                                                 for (k=1; k<=d; k++ ) {                                                 
00277                                                         _cluster_centroid[_clusters[i] ][k] += entries[j][k];
00278                                                 }
00279                                                 ++count[ _clusters[i] ];
00280                                                 break;
00281                                         }
00282                         }
00283                 }
00284         }
00285 
00286         for (i=1; i<= _num_clusters; i++) for (k=1; k<=d; k++ ) _cluster_centroid[i][k] /= count[i];
00287 
00288                 // Calculo da distancia M entre os centroides ij
00289         Value_Type sum;
00290         for (i=1; i<= _num_clusters; i++)
00291                 for (j=1; j<= _num_clusters; j++) {
00292                         sum = 0.0;
00293                         for (k=1; k<= d; k++) sum += pow( fabs(_cluster_centroid[i][k] - _cluster_centroid[j][k]), p );
00294                         _M[i][j] = pow( sum, (1.0/p) );                 
00295                 }
00296 
00297                 // Calculo da dispers�o S nos agrupamentos i
00298         for (i=1; i<=_num_clusters; i++) count[i] = 0.0;
00299 
00300         for (ii=1; ii<= _num_clusters; ii++)
00301         for (i=1; i<=m; i++ ) {
00302                 if ( _clusters[i] == ii ) {
00303                         for (l=0; l< codebook_label[i-1].size(); l++ )
00304                                 for (j=1; j<=n; j++) {
00305                                         if ( codebook_label[i-1][l] == entries_label[j-1] ) {
00306                                                 sum = 0.0;
00307                                                 for (k=1; k<=d; k++ ) {                                                 
00308                                                         sum += pow( fabs( entries[j][k] - _cluster_centroid[ii][k]), q );
00309                                                 }
00310                                                 _S[ii] += sum;
00311                                                 count[ii]++;
00312                                                 break;
00313                                         }
00314                         }
00315                 }
00316         }
00317 /*
00318         for (i=1; i<= _num_clusters; i++)
00319                 for (j=1; j<=m; j++ ) {
00320                         if ( _clusters[j] == i  ) {
00321                                 count[i]++;
00322                                 sum = 0.0;
00323                                 for (k=1; k<=d; k++) sum += pow( fabs( codebook[j][k] - _cluster_centroid[i][k]), q );                                                          
00324                                 _S[i] += sum;
00325                         }
00326                 }
00327 */
00328         for (i=1; i<= _num_clusters; i++)   _S[i] = pow( (_S[i]/count[i]) ,(1.0/q) );
00329         
00330 
00331 
00332         //printmatrix( _M, 1, _num_clusters,1,_num_clusters);
00333         //cout << endl;
00334         //for (i=1; i<= _num_clusters; i++ ) cout << _S[i] << " "; cout << endl;
00335 
00336         for (i=1; i<= _num_clusters; i++)
00337                 for (j=1; j<= _num_clusters; j++)
00338                         if ( i!=j )
00339                                 _R[i][j] = ( _S[i] + _S[j] )/ _M[i][j];
00340                         else
00341                                 _R[i][j] = 0;   
00342 
00343         //printmatrix( _R,1, _num_clusters,1,_num_clusters);
00344 
00345         Value_Type result = 0.0, max;
00346         for (i=1; i<= _num_clusters; i++) {
00347                 max = 0.0; 
00348                 for (j=1; j<= _num_clusters; j++) {
00349                         if (_R[i][j] > max)
00350                                 max = _R[i][j];
00351                 }
00352                 result += max;
00353         }
00354         
00355         result /= (Value_Type)_num_clusters;
00356 
00357         return result;
00358 };  
00359 
00360 Value_Type  
00361 GraphSegmentation::Davies_Bouldin( Value_Type p, Value_Type q )
00362 {
00363         int i, j, k;
00364         int d                           =  _som->getMapcode()->getNumVar();
00365         TMatrix codebook        =  _som->getMapcode()->getCodebook();
00366         
00367 
00368         TVector         _S                                      = create_vector( 1, _num_clusters );
00369         TMatrix         _cluster_centroid       = create_matrix(1, _num_clusters, 1, d );
00370         TIntVector      _num_vector_cluster = create_intvector( 1, _num_clusters );
00371         TMatrix         _M                                      = create_matrix( 1, _num_clusters,1, _num_clusters );
00372         TMatrix         _R                                      = create_matrix( 1, _num_clusters,1, _num_clusters );
00373 
00374         for (i=1; i<= _num_clusters; i++ ) {
00375                 _S[i] = 0.0; 
00376                 _num_vector_cluster[i] = 0;
00377                 for (j=1; j <= _num_clusters; j++ )     _R[i][j] = _M[i][j] = 0.0;
00378                 for (j=1; j<= d; j++ )  _cluster_centroid[i][j] = 0.0;
00379         }
00380 
00381         // Calculo dos centroides <<Vetores de C�digo>>
00382         TVector count = create_vector( 1, _num_clusters ); for (i=1;i<=_num_clusters; i++) count[i] = 0.0;
00383         for (i=1; i<=m; i++) 
00384                 if ( _clusters[i] > 0 ) 
00385                 {       
00386                         for (k=1;k<=d;k++) 
00387                         {
00388                                 _cluster_centroid[_clusters[i] ][k] += codebook[i][k];          
00389                         }
00390                         ++count[ _clusters[i] ];
00391                 }
00392         for (i=1; i<= _num_clusters; i++) for (k=1; k<=d; k++ ) _cluster_centroid[i][k] /= count[i];
00393         
00394         // Calculo da distancia M entre os centroides ij
00395         Value_Type sum;
00396         for (i=1; i<= _num_clusters; i++)
00397                 for (j=1; j<= _num_clusters; j++) {
00398                         sum = 0.0;
00399                         for (k=1; k<= d; k++) sum += pow( fabs(_cluster_centroid[i][k] - _cluster_centroid[j][k]), p );
00400                         _M[i][j] = pow( sum, (1.0/p) );                 
00401                 }
00402 
00403         
00404         // Calculo da dispers�o S nos agrupamentos i
00405         for (i=1; i<=_num_clusters; i++) count[i] = 0.0;
00406 
00407         for (i=1; i<= _num_clusters; i++)
00408                 for (j=1; j<=m; j++ ) {
00409                         if ( _clusters[j] == i  ) {
00410                                 count[i]++;
00411                                 sum = 0.0;
00412                                 for (k=1; k<=d; k++) sum += pow( fabs( codebook[j][k] - _cluster_centroid[i][k]), q );                                                          
00413                                 _S[i] += sum;
00414                         }
00415                 }
00416         for (i=1; i<= _num_clusters; i++)   _S[i] = pow( (_S[i]/count[i]) ,(1.0/q) );
00417         
00418 
00419 
00420         //printmatrix( _M, 1, _num_clusters,1,_num_clusters);
00421         //cout << endl;
00422         //for (i=1; i<= _num_clusters; i++ ) cout << _S[i] << " "; cout << endl;
00423 
00424         for (i=1; i<= _num_clusters; i++)
00425                 for (j=1; j<= _num_clusters; j++)
00426                         if ( i!=j )
00427                                 _R[i][j] = ( _S[i] + _S[j] )/ _M[i][j];
00428                         else
00429                                 _R[i][j] = 0;   
00430 
00431         //printmatrix( _R,1, _num_clusters,1,_num_clusters);
00432 
00433         Value_Type result = 0.0, max;
00434         for (i=1; i<= _num_clusters; i++) {
00435                 max = 0.0; 
00436                 for (j=1; j<= _num_clusters; j++) {
00437                         if (_R[i][j] > max)
00438                                 max = _R[i][j];
00439                 }
00440                 result += max;
00441         }
00442         
00443         result /= (Value_Type)_num_clusters;
00444 
00445         return result;
00446 };  
00447 
00448 
00449 void 
00450 GraphSegmentation::Init( SOM* som ) 
00451 {
00452         _som          = som;
00453         _adjacentNode = AdjacentNode();         
00454         _distance     = Distance(  );
00455         _winner           = Winner( );
00456         _meanDistance = MeanDistance( );
00457         _centroid         = Centroid(  );
00458         _Hmin             = Hmin( );
00459         m                         = _som->getMapcode()->getMapSize();
00460         _delta            = 0.5;
00461         
00462 }
00463 
00464 void  
00465 GraphSegmentation::Label( int i, int& group, TIntVector& vec, bool & changed ) {
00466         int j;
00467         
00468         for (j=1; j<=m; j++) 
00469                 if ((_adjacentNode[i][j] == 1) && (vec[j] == 0)) {
00470                         changed = true;
00471                         _adjacentNode[i][j] = 0;
00472                         vec[j] = group;
00473                         if ( i != j )
00474                                 Label( j, group, vec, changed );                        
00475                 }
00476 };
00477 
00478 TIntVector 
00479 GraphSegmentation::Segment( SOM* som, int min_neurons_per_cluster ) {
00480 
00481         Init( som );
00482         int i, j;
00483 
00484         for (i=1; i<=m; i++) 
00485                 for (j=1; j<=m; j++) 
00486                         if ( _adjacentNode[i][j] == 1 ) {
00487                                 if (( _distance[i][j] > 2*_meanDistance[i]) ||
00488                                         ( _distance[i][j] > 2*_meanDistance[j]) )
00489                                         _adjacentNode[i][j] = 0;
00490                                 if ((_winner[i] == 0) || 
00491                                         (_winner[j] == 0) || 
00492                                         (( _winner[i] < _Hmin ) && (_winner[j] < _Hmin )) )
00493                                         _adjacentNode[i][j] = 0;
00494                                 if ( _centroid[i][j] > 2*_distance[i][j] )
00495                                         _adjacentNode[i][j] = 0;
00496                         }       
00497 
00498         TIntVector tmp = create_intvector( 1, m ); for (i=1;i<=m;i++) tmp[i] = 0;
00499         int group = 1;
00500         bool changed = false;
00501 
00502         for ( i=1; i<= m; i++ )  {
00503                 Label( i, group, tmp, changed );
00504                 if ( changed ) {
00505                         group++;
00506                         changed = false;
00507                 }
00508         };
00509         
00510         int MaxNumCluster = 0;
00511         for (i=1; i<=m; i++) if (tmp[i] > MaxNumCluster) MaxNumCluster = i;
00512 
00513         TIntVector tmp_clusters = create_intvector( 0, MaxNumCluster ); for (i=0;i<=MaxNumCluster;i++) tmp_clusters[i] = 0;
00514         for (i=1; i<=m; i++) tmp_clusters[tmp[i]]++;
00515         for (i=1; i<=m; i++) if ( (tmp[i]>0) && ( tmp_clusters[tmp[i]] < min_neurons_per_cluster ) )
00516         { 
00517                 tmp[i] = 0;
00518         };
00519 
00520 
00521         vector<int> vec, vec2;
00522         for (i=1;i<=m;i++) vec.push_back( tmp[i] );
00523         sort( vec.begin(), vec.end() );
00524         
00525         int tmp_n = vec[0];
00526         for (i=0;i<m;i++) if (vec[i] != tmp_n) 
00527                                                         {
00528                                                                 tmp_n = vec[i];                                                                                                                 
00529                                                                 vec2.push_back( vec[i] );
00530                                                         }
00531         _num_clusters = vec2.size();
00532         for (i=1;i<=m;i++) 
00533                 for (j=0; j< _num_clusters; j++ )
00534                         if ( tmp[i] == vec2[j] )
00535                                 tmp[i] = j+1;
00536         
00537         //cout << _num_clusters << endl;
00538 
00539         free_intvector( tmp_clusters, 0, MaxNumCluster );
00540         
00541         _clusters = tmp;
00542         return tmp;
00543 };
00544 
00545 TMatrix 
00546 GraphSegmentation::Centroid(  ) {
00547         int m                           =  _som->getMapcode()->getMapSize();
00548         int n                           =  _som->getData()->getDataSize();
00549         int d                           =  _som->getMapcode()->getNumVar();
00550         int c;
00551         TMatrix entries         =  _som->getData()->getEntries();
00552         TMatrix codebook        =  _som->getMapcode()->getCodebook();
00553         int i, j;
00554 
00555         TMatrix centroids = create_matrix( 1, m, 1, d );
00556         for (i=1; i<=m; i++) for (j=1; j<=d; j++) centroids[i][j] = 0.0;
00557         TVector count     = create_vector( 1, m );
00558         for (j=1; j<=m; j++) count[j] = 0;
00559 
00560         Value_Type min_dist;
00561 
00562         for (i=1; i<=n; i++)    {       
00563                 min_dist = 9999999;
00564                 for (j=1; j<=m; j++) 
00565                         if ( dist( entries[i], codebook[j], d ) < min_dist ) {
00566                                 min_dist = dist( entries[i], codebook[j], d );
00567                                 c = j;
00568                         }
00569                 for (j=1; j<=d; j++) centroids[c][j] += entries[i][j];
00570                 count[c]++;
00571         }
00572 
00573         for (i=1; i<=m; i++) for (j=1; j<=d; j++)
00574                 if ( count[i] != 0 )
00575                         centroids[i][j] /= count[i];
00576 
00577         TMatrix dist_centroids = create_matrix( 1, m, 1, m );
00578         for (i=1; i<=m; i++) for (j=1; j<=m; j++) dist_centroids[i][j] = 0.0;
00579         
00580         for (i=1; i<=m; i++) for (j=1; j<=m; j++)
00581                 dist_centroids[i][j] = dist( centroids[i], centroids[j], d );
00582 
00583         return dist_centroids;
00584 };
00585 
00586 TMatrix 
00587 GraphSegmentation::Distance(  ) {
00588         int m                           =  _som->getMapcode()->getMapSize();
00589         int d                           =  _som->getMapcode()->getNumVar();
00590         TMatrix codebook        =  _som->getMapcode()->getCodebook();
00591         int i, j;
00592 
00593         TMatrix tmp = create_matrix( 1, m, 1, m );
00594         
00595         for (i=1; i<=m; i++)
00596                 for (j=1; j<=m; j++)
00597                         tmp[i][j] = dist( codebook[i], codebook[j], d );
00598         
00599         return tmp;
00600 };
00601 
00602 TVector 
00603 GraphSegmentation::Winner( ) {
00604         int m                   =  _som->getMapcode()->getMapSize();
00605         TLabel maplabel =  _som->getMapcode()->getLabel();
00606 
00607         TVector tmp = create_vector( 1, m );
00608         int count = 1;
00609         for (TLabel::iterator i = maplabel.begin(); i != maplabel.end(); i++ ) {
00610                 if ( (*i).empty() )
00611                         tmp[count] = 0;
00612                 else
00613                         tmp[count] = (*i).size();
00614                 count++;
00615         }
00616         return tmp;
00617 };
00618 
00619 TVector 
00620 GraphSegmentation::MeanDistance( ) {
00621         int m                   =  _som->getMapcode()->getMapSize();
00622         int i, j;
00623         int d                           =  _som->getMapcode()->getNumVar();
00624         TMatrix codebook        =  _som->getMapcode()->getCodebook();
00625 
00626         TVector mean_distance = create_vector(1,m );
00627         for (i=1; i<=m; i++) mean_distance[i] = 0;
00628         TVector count = create_vector(1, m );
00629         for (j=1; j<=m; j++) count[j] = 0;
00630 
00631         for (i=1; i<= m; i++)
00632                 for (j=1; j<= m; j++)
00633                         if ( ( _adjacentNode[i][j] == 1 ) && ( i != j ) ) {                             
00634                                 mean_distance[i] += dist( codebook[j], codebook[i], d );
00635                                 count[i]++;
00636                         }
00637 
00638         for (i=1; i<=m; i++) 
00639                 mean_distance[i] /= count[i];                   
00640 
00641         return mean_distance;
00642 };
00643 
00644 TMatrix 
00645 GraphSegmentation::AdjacentNode() {
00646         TMatrix delta   =  _som->getDelta();
00647         int m                   =  _som->getMapcode()->getMapSize();
00648         int i, j;
00649         
00650         TMatrix tmp = create_matrix( 1, m, 1, m );      
00651 
00652         for (i=1; i<=m; i++)
00653                 for (j=1; j<=m; j++)
00654                         if ( delta[i][j] <= sqrt(2) )
00655                                 tmp[i][j] = 1;
00656                         else
00657                                 tmp[i][j] = 0;
00658 
00659         return tmp;
00660 };
00661 
00662 Value_Type    
00663 GraphSegmentation::Hmin( ) {
00664         int m   =  _som->getMapcode()->getMapSize();
00665         int n   =  _som->getData()->getDataSize();
00666         
00667         return ( ((Value_Type)m/(Value_Type)n)*_delta );
00668 };
00669 
00670 #endif