cluster.cpp

   1 /*
   2  *  cluster.cpp
   3  *
   4  *
   5  *  Created by Pat Schloss on 8/14/08.
   6  *  Copyright 2008 Patrick D. Schloss. All rights reserved.
   7  *
   8  */
   9
  10 #include "cluster.hpp"
  11 #include "rabundvector.hpp"
  12 #include "listvector.hpp"
  13 #include "sparsematrix.hpp"
  14
  15 /***********************************************************************/
  16
  17 Cluster::Cluster(RAbundVector* rav, ListVector* lv, SparseMatrix* dm, float c) :
  18 rabund(rav), list(lv), dMatrix(dm)
  19 {
  20 /*
  21         cout << "sizeof(MatData): " << sizeof(MatData) << endl;
  22         cout << "sizeof(PCell*): " << sizeof(PCell*) << endl;
  23
  24         int nCells = dMatrix->getNNodes();
  25         time_t start = time(NULL);
  26
  27         MatVec matvec = MatVec(nCells);
  28         int i = 0;
  29         for (MatData currentCell = dMatrix->begin(); currentCell != dMatrix->end(); currentCell++) {
  30                 matvec[i++] = currentCell;
  31         }
  32         for (i= matvec.size();i>0;i--) {
  33                 dMatrix->rmCell(matvec[i-1]);
  34         }
  35         MatData it = dMatrix->begin();
  36         while (it != dMatrix->end()) {
  37                 it = dMatrix->rmCell(it);
  38         }
  39         cout << "Time to remove " << nCells << " cells: " << time(NULL) - start << " seconds" << endl;
  40     exit(0);
  41         MatData it = dMatrix->begin();
  42         cout << it->row << "/" << it->column << "/" << it->dist << endl;
  43         dMatrix->rmCell(dMatrix->begin());
  44         cout << it->row << "/" << it->column << "/" << it->dist << endl;
  45         exit(0);
  46 */
  47
  48         // Create a data structure to quickly access the PCell information
  49         // for a certain sequence. It consists of a vector of lists, where
  50         // a list contains pointers (iterators) to the all distances related
  51         // to a certain sequence. The Vector is accessed via the index of a
  52         // sequence in the distance matrix.
  53         seqVec = vector<MatVec>(lv->size());
  54         for (MatData currentCell = dMatrix->begin(); currentCell != dMatrix->end(); currentCell++) {
  55                 seqVec[currentCell->row].push_back(currentCell);
  56                 seqVec[currentCell->column].push_back(currentCell);
  57         }
  58         mapWanted = false;  //set to true by mgcluster to speed up overlap merge
  59
  60         //save so you can modify as it changes in average neighbor
  61         cutoff = c;
  62 }
  63
  64 /***********************************************************************/
  65
  66 void Cluster::getRowColCells() {
  67         try {
  68                 PCell* smallCell = dMatrix->getSmallestCell();  //find the smallest cell - this routine should probably not be in the SpMat class
  69
  70                 smallRow = smallCell->row;              // get its row
  71                 smallCol = smallCell->column;   // get its column
  72                 smallDist = smallCell->dist;    // get the smallest distance
  73
  74                 rowCells = seqVec[smallRow];    // all distances related to the row index
  75                 colCells = seqVec[smallCol];    // all distances related to the column index
  76                 nRowCells = rowCells.size();
  77                 nColCells = colCells.size();
  78         }
  79         catch(exception& e) {
  80                 errorOut(e, "Cluster", "getRowColCells");
  81                 exit(1);
  82         }
  83
  84 }
  85 /***********************************************************************/
  86 // Remove the specified cell from the seqVec and from the sparse
  87 // matrix
  88 void Cluster::removeCell(const MatData& cell, int vrow, int vcol, bool rmMatrix)
  89 {
  90         ull drow = cell->row;
  91         ull dcol = cell->column;
  92         if (((vrow >=0) && (drow != smallRow)) ||
  93                 ((vcol >=0) && (dcol != smallCol))) {
  94                 ull dtemp = drow;
  95                 drow = dcol;
  96                 dcol = dtemp;
  97         }
  98
  99         ull crow;
 100         ull ccol;
 101         int nCells;
 102         if (vrow < 0) {
 103                 nCells = seqVec[drow].size();
 104                 for (vrow=0; vrow<nCells;vrow++) {
 105                         crow = seqVec[drow][vrow]->row;
 106                         ccol = seqVec[drow][vrow]->column;
 107                         if (((crow == drow) && (ccol == dcol)) ||
 108                                 ((ccol == drow) && (crow == dcol))) {
 109                                 break;
 110                         }
 111                 }
 112         }
 113         seqVec[drow].erase(seqVec[drow].begin()+vrow);
 114         if (vcol < 0) {
 115                 nCells = seqVec[dcol].size();
 116                 for (vcol=0; vcol<nCells;vcol++) {
 117                         crow = seqVec[dcol][vcol]->row;
 118                         ccol = seqVec[dcol][vcol]->column;
 119                         if (((crow == drow) && (ccol == dcol)) ||
 120                                 ((ccol == drow) && (crow == dcol))) {
 121                                 break;
 122                         }
 123                 }
 124         }
 125         seqVec[dcol].erase(seqVec[dcol].begin()+vcol);
 126         if (rmMatrix) {
 127                 dMatrix->rmCell(cell);
 128         }
 129 }
 130
 131
 132 /***********************************************************************/
 133
 134 void Cluster::clusterBins(){
 135         try {
 136         //      cout << smallCol << '\t' << smallRow << '\t' << smallDist << '\t' << rabund->get(smallRow) << '\t' << rabund->get(smallCol);
 137
 138                 rabund->set(smallCol, rabund->get(smallRow)+rabund->get(smallCol));
 139                 rabund->set(smallRow, 0);
 140                 rabund->setLabel(toString(smallDist));
 141
 142         //      cout << '\t' << rabund->get(smallRow) << '\t' << rabund->get(smallCol) << endl;
 143         }
 144         catch(exception& e) {
 145                 errorOut(e, "Cluster", "clusterBins");
 146                 exit(1);
 147         }
 148
 149
 150 }
 151
 152 /***********************************************************************/
 153
 154 void Cluster::clusterNames(){
 155         try {
 156         //      cout << smallCol << '\t' << smallRow << '\t' << smallDist << '\t' << list->get(smallRow) << '\t' << list->get(smallCol);
 157                 if (mapWanted) {  updateMap();  }
 158
 159                 list->set(smallCol, list->get(smallRow)+','+list->get(smallCol));
 160                 list->set(smallRow, "");
 161                 list->setLabel(toString(smallDist));
 162
 163         //      cout << '\t' << list->get(smallRow) << '\t' << list->get(smallCol) << endl;
 164     }
 165         catch(exception& e) {
 166                 errorOut(e, "Cluster", "clusterNames");
 167                 exit(1);
 168         }
 169
 170 }
 171
 172 /***********************************************************************/
 173 //This function clusters based on the method of the derived class
 174 //At the moment only average and complete linkage are covered, because
 175 //single linkage uses a different approach.
 176 void Cluster::update(double& cutOFF){
 177         try {
 178                 getRowColCells();
 179
 180                 vector<int> foundCol(nColCells, 0);
 181
 182                 int search;
 183                 bool changed;
 184
 185                 // The vector has to be traversed in reverse order to preserve the index
 186                 // for faster removal in removeCell()
 187                 for (int i=nRowCells-1;i>=0;i--) {
 188                         if (!((rowCells[i]->row == smallRow) && (rowCells[i]->column == smallCol))) {
 189                                 if (rowCells[i]->row == smallRow) {
 190                                         search = rowCells[i]->column;
 191                                 } else {
 192                                         search = rowCells[i]->row;
 193                                 }
 194
 195                                 bool merged = false;
 196                                 for (int j=0;j<nColCells;j++) {
 197                                         if (!((colCells[j]->row == smallRow) && (colCells[j]->column == smallCol))) { //if you are not hte smallest distance
 198                                                 if (colCells[j]->row == search || colCells[j]->column == search) {
 199                                                         foundCol[j] = 1;
 200                                                         merged = true;
 201                                                         changed = updateDistance(colCells[j], rowCells[i]);
 202                                                         // If the cell's distance changed and it had the same distance as
 203                                                         // the smallest distance, invalidate the mins vector in SparseMatrix
 204                                                         if (changed) {
 205                                                                 if (colCells[j]->vectorMap != NULL) {
 206                                                                         *(colCells[j]->vectorMap) = NULL;
 207                                                                         colCells[j]->vectorMap = NULL;
 208                                                                 }
 209                                                         }
 210                                                         break;
 211                                                 }
 212                                         }
 213                                 }
 214                                 //if not merged it you need it for warning
 215                                 if (!merged) {
 216                                         mothurOut("Warning: trying to merge cell " + toString(rowCells[i]->row+1) + " " + toString(rowCells[i]->column+1) + " distance " + toString(rowCells[i]->dist) + " with value above cutoff. Results may vary from using cutoff at cluster command instead of read.dist."); mothurOutEndLine();
 217                                         if (cutOFF > rowCells[i]->dist) {  cutOFF = rowCells[i]->dist;  mothurOut("changing cutoff to " + toString(cutOFF));  mothurOutEndLine(); }
 218
 219                                 }
 220                                 removeCell(rowCells[i], i , -1);
 221
 222                         }
 223                 }
 224                 clusterBins();
 225                 clusterNames();
 226
 227                 // Special handling for singlelinkage case, not sure whether this
 228                 // could be avoided
 229                 for (int i=nColCells-1;i>=0;i--) {
 230                         if (foundCol[i] == 0) {
 231                                 if (!((colCells[i]->row == smallRow) && (colCells[i]->column == smallCol))) {
 232                                         mothurOut("Warning: merging cell " + toString(colCells[i]->row+1) + " " + toString(colCells[i]->column+1) + " distance " + toString(colCells[i]->dist) + " value above cutoff. Results may vary from using cutoff at cluster command instead of read.dist."); mothurOutEndLine();
 233                                         if (cutOFF > colCells[i]->dist) {  cutOFF = colCells[i]->dist;  mothurOut("changing cutoff to " + toString(cutOFF));  mothurOutEndLine(); }
 234                                 }
 235                                 removeCell(colCells[i], -1, i);
 236                         }
 237                 }
 238         }
 239         catch(exception& e) {
 240                 errorOut(e, "Cluster", "update");
 241                 exit(1);
 242         }
 243 }
 244 /***********************************************************************/
 245 void Cluster::setMapWanted(bool m)  {
 246         try {
 247                 mapWanted = m;
 248
 249                 //initialize map
 250                 for (int i = 0; i < list->getNumBins(); i++) {
 251
 252                         //parse bin
 253                         string names = list->get(i);
 254                         while (names.find_first_of(',') != -1) {
 255                                 //get name from bin
 256                                 string name = names.substr(0,names.find_first_of(','));
 257                                 //save name and bin number
 258                                 seq2Bin[name] = i;
 259                                 names = names.substr(names.find_first_of(',')+1, names.length());
 260                         }
 261
 262                         //get last name
 263                         seq2Bin[names] = i;
 264                 }
 265
 266         }
 267         catch(exception& e) {
 268                 errorOut(e, "Cluster", "setMapWanted");
 269                 exit(1);
 270         }
 271 }
 272 /***********************************************************************/
 273 void Cluster::updateMap() {
 274 try {
 275                 //update location of seqs in smallRow since they move to smallCol now
 276                 string names = list->get(smallRow);
 277                 while (names.find_first_of(',') != -1) {
 278                         //get name from bin
 279                         string name = names.substr(0,names.find_first_of(','));
 280                         //save name and bin number
 281                         seq2Bin[name] = smallCol;
 282                         names = names.substr(names.find_first_of(',')+1, names.length());
 283                 }
 284
 285                 //get last name
 286                 seq2Bin[names] = smallCol;
 287
 288         }
 289         catch(exception& e) {
 290                 errorOut(e, "Cluster", "updateMap");
 291                 exit(1);
 292         }
 293 }
 294 /***********************************************************************/
 295
 296
 297