https://github.com/cran/dtw
Raw File
Tip revision: 9fbfbef539cfc60884f6828c62f0aa328335a0b1 authored by Toni Giorgino on 08 January 2008, 00:00:00 UTC
version 1.5-3
Tip revision: 9fbfbef
globalCostMatrix.R
###############################################################
#                                                             #
#   (c) Toni Giorgino <toni.giorgino@gmail.com>               #
#       Laboratory for Biomedical Informatics                 #
#       University of Pavia - Italy                           #
#       www.labmedinfo.org                                    #
#                                                             #
#   $Id: globalCostMatrix.R 83 2008-01-04 00:25:00Z tonig $
#                                                             #
###############################################################


########################################
## Compute the cost matrix from a local distance matrix

## Wrapper to the native function





`globalCostMatrix` <-
function(lm,
         step.matrix=symmetric1,
         window.function=noWindow,
         native=TRUE,
         ...) {


  ## sanity check - be extra cautions w/ binary
  if (!is.stepPattern(step.matrix))
    stop("step.matrix is no stepMatrix object");



  # i = 1 .. n in query sequence, on first index, ie rows
  # j = 1 .. m on template sequence, on second index, ie columns
  #   Note:  template is usually drawn vertically, up-wise

  n <- nrow(lm);
  m <- ncol(lm);


  # number of individual steps (counting all patterns)
  nsteps<-dim(step.matrix)[1];


  # clear the cost and step matrix
  # these will be the outputs of the binary
  cm <- matrix(NA,nrow=n,ncol=m);
  sm <- matrix(NA,nrow=n,ncol=m);


  
  # initializer
  cm[1,1] <- lm[1,1];

  if(is.loaded("computeCM") && native){
    ## precompute windowing
    wm <- matrix(FALSE,nrow=n,ncol=m);
    wm[window.function(row(wm),col(wm),
                       query.size=n, template.size=m,
                       ...)]<-TRUE;

    ## this call could be optimized
    tmp<-.C("computeCM",NAOK=TRUE,PACKAGE="dtw",
            as.integer(dim(cm)),               # s
            as.logical(wm),                    #
            as.double(lm),
            as.integer(nsteps),
            as.double(step.matrix),
            cmo=as.double(cm),                     # OUT
            smo=as.integer(sm));                   # OUT

    cm<-matrix(tmp$cmo,nrow=n,ncol=m);
    sm<-matrix(tmp$smo,nrow=n,ncol=m);

  } else {

    ####################
    ## INTERPRETED PURE-R IMPLEMENTATION
    
    warning("Native dtw implementation not available: using (slow) interpreted fallback");
                                        # now walk through the matrix, column-wise and row-wise,
                                        # and recursively compute the accumulated distance. Unreachable
                                        # elements are handled via NAs (removed)
    dir <- step.matrix;
    npats <- attr(dir,"npat");
    for (j in 1:m) {
      for (i in 1:n) {
        ## It is ok to window on the arrival point (?)
        if(!window.function(i,j, query.size=n, template.size=m, ...)) { next; }

        clist<-numeric(npats)+NA;
        for (s in 1:nsteps) {
          ## current pattern
          p<-dir[s,1];
          ## ii,jj is the cell from where potentially we could
          ## have come from. 
          ii<-i-dir[s,2];                 # previous step in inp
          jj<-j-dir[s,3];                 # previous step in tpl
          if(ii>=1 && jj>=1) {            # element exists?
            cc<-  dir[s,4];               # step penalty
            if(cc == -1) {		#  -1? cumulative cost:
              clist[p]<-cm[ii,jj];	#  there must be exactly 1 per pattern
            } else {			#  a cost for 
              clist[p]<-clist[p]+cc*lm[ii,jj];
            }
          }
        }


        ## no NAs in clist at this point BUT clist can be empty
        ## store in cost matrix
        minc<-which.min(clist);           # pick the least cost
        if(length(minc)>0) {          	# false if clist has all NAs
          cm[i,j]<-clist[minc];
          sm[i,j]<-minc;			# remember the pattern picked
        }
      }
    }
  }

  ## END PURE-R IMPLEMENTATION
  ####################


  out<-list();
  out$costMatrix<-cm;                   # to get distance
  out$directionMatrix<-sm;              # to backtrack
  out$stepPatterns<-step.matrix;        # to backtrack

  return(out);
}

back to top