alexdepremia / advanced-density-peaks Goto Github PK

Sometimes, the dendrogram generation goes in an infinite loop, generating a huge gpl file. Probably we should program again the subroutine graph2 (and simplify it).

I want to use other datasets. How can I get the file with the suffix ".in" in the folder "examples"?
Thank you for your help!

I'm unsure about the part of the code in which borders between clusters are identified (lines 1978-2013)

#!fortran90

do i=1,Nele
  if (.not.centquest(i)) then
    dmin=maxr
    do j=1,Nele
      if (cluster(j).ne.cluster(i)) then
        d=gDist(i,j)
        if (d.lt.dmin) then
          dmin=d
          jg=j
        endif
      endif
    enddo
    if (dmin.le.dc(i)) then
      iref=i
      k=0
      extend=.true.
      do while ( (k.lt.Nele).and.extend)
        k=k+1
        if (cluster(k).eq.cluster(i)) then
          if (gDist(k,jg).lt.dmin) extend=.false.
        endif
      enddo
      if (extend) then
        candidates_B(i)=cluster(jg)
        if (Rho(iref).gt. Bord(cluster(i),cluster(jg))) then
          Bord(cluster(i),cluster(jg))=Rho(iref)
          Bord(cluster(jg),cluster(i))=Rho(iref)
          Bord_err(cluster(i),cluster(jg))=Rho_err(iref)
          Bord_err(cluster(jg),cluster(i))=Rho_err(iref)
          eb(cluster(i),cluster(jg))=iref
          eb(cluster(jg),cluster(i))=iref
        endif
      endif
    endif
  endif
enddo

Specifically, the first inner loop

#!fortran90

do j=1,Nele
      if (cluster(j).ne.cluster(i)) then
        d=gDist(i,j)
        if (d.lt.dmin) then
          dmin=d
          jg=j
        endif
      endif
    enddo

given a point i, will find its nearest neighbour not belonging to cluster(i), which will be labelled ig. Then, the following step is to check if there's any point in cluster(i) closer to ig.
This is similar but not equal to what I read in the arxiv manuscript:

"A point i, belonging to cluster c, is assumed to be at the border between cluster c and c' if its closest point j belonging to c' is within a distance r_k_i and if i is the closest point to j among those belonging to c"

In particular, according to the manuscript definition, given cluster c, and point i, one should consider each of the other clusters, and find the point closest to i for each of the other clusters, then do the second check on each of them.

Am I interpreting any of this wrong?
Sorry if there's something obvious that I'm missing