[R-sig-eco] some MDS and R problems
Gavin Simpson
gavin.simpson at ucl.ac.uk
Tue Jun 15 19:17:46 CEST 2010
On Tue, 2010-06-15 at 20:00 +0300, Jari Oksanen wrote:
> On 15/06/10 15:15 PM, "Gavin Simpson" <gavin.simpson at ucl.ac.uk> wrote:
>
> > Hi Arnaldo,
> >
> > You ask several disjointed question that I don't really follow (esp your
> > last point about unique), but in response to:
> >
> >> When I delete a sample I do not loose some weight?
> >
> > No, nMDS only cares about trying to position points in a k dimensional
> > space such that the distance between points in this k dimensional space
> > best represents the rank ordering of the original dissimilarities. If
> > two samples have 0 dissimilarity, they should occupy the same location
> > on the plot (in the k dimensional space); one does not influence the
> > other. Hence simply deleting one of the pair of samples that have zero
> > distance to one another is appropriate.
> >
> Actually, you *do* lose weights. If two points are identical, all other
> points have twice the same contribution to the stress with respect to this
> identical pair. There is no way of handling this in isoMDS() which knows no
> weights and does not allow duplicate points . The 'zerodist' trick in vegan
> metaMDS() is just a kluge to get around this.
Ahh, yes. I wasn't thinking about things that way. Taking a step back,
nMDS is about finding a mapping in some low dimensional space. If two
points are exactly similar they should be plotted in the the same
location. In that sense, you don't loose anything by dropping one out,
in the sense that the end result is the location for your two identical
points. That was what I was referring to.
G
> The 'zerodist' argument is not used in metaMDS() if you supply your own
> dissimilarities. The 'zerodist' works within metaMDSdist() that calculates
> dissimilaritiesa and it is not called if you supply your own
> dissimilarities. This is not documented in vegan explicitly (I'll see how to
> do this). If you have your own dissimilarity structure, you should handle
> zero distances by hand before calling metaMDS() or isoMDS().
>
> cheers, Jari Oksanen
>
> > As for the code I wrote in the email you cite:
> >
> > minDij <- min(Dij[Dij > 0]) / 2
> > Dij[Dij <= 0] <- minDij
> >
> > We assume Dij is the object containing your distance matrix. In your
> > example code, it would be
> >
> > Dij <- as.matrix(dissim1)
> >
> > for example.
> >
> > The first line of my code calculates half the smallest observed
> > dissimilarity. If we break it down, the code in that first line does the
> > following [there was a missing `]` in the version you quote and probably
> > in my original]:
> >
> > Dij[Dij > 0] subsets Dij so that we only work with the dissimilarities
> > that are positive. This way of indexing the matrix Dij results in a
> > vector of positive dissimilarities. We then apply the min( ) function to
> > this vector of positive dissimilarities to return the smallest positive
> > dissimilarity observed. We then halve this minimum value and store it in
> > the object minDij.
> >
> > The second line of my code uses this dissimilarity value (minDij) to
> > replace the observed dissimilarities in Dij that are less than or equal
> > to 0.
> >
> > Hope this is of use. If you still can't get this to work, post back and
> > let us know.
> >
> > All the best,
> >
> > G
> >
> > On Tue, 2010-06-15 at 08:34 -0300, Arnaldo Russo wrote:
> >> Hi all,
> >>
> >> I'm executing some multivariate exploratory annalysis, with vegan and MASS
> >> packages.
> >>
> >> Some problems are not solved for me at this moment. I have read all past
> >> discussions over MDS problems.
> >>
> >> In my studies occurs a situation that some samples are equal each other, and
> >> this results in zero dissimilarities.
> >>
> >> I tryed with no success, use the stepacross function. So, in this case I
> >> changed one of these zero dissimilarities values for a minimum (1e-5). Some
> >> of these modifications are different than "stepacross" or same isoMDS(x.dist
> >> + 1e-5) (proposed by Jari oksanen to 'lie' to isoMDS? When I delete a
> >> sample I do not loose some weight?
> >>
> >> Changing one of those zero dissimilarities and using zerodist = "add" option
> >> of metaMDS, it passed the current error (zero or negative distance between
> >> objects).
> >>
> >> Someone could explain the function posted by Gavin Simpson (Jan 12, 2010;
> >> 05:45pm Re: Non-metric multidimensional scaling (NMDS) help). I didn't get
> >> some result.
> >>
> >> If there is a good reason, and you want to include all samples, then
> >>>
> >> you'll need to come up with a means for handling them. metaMDSdist allow
> >>>
> >> you to add a small value to the zero dissimilarities. The details are in
> >>> the code, but effectively all zero distances are replaced by half the
> >>> smallest non zero distance. You could do a similar replacement yourself
> >>> if you feel this is warranted and/or justified.
> >>>
> >>> minDij <- min(Dij[Dij > 0) / 2
> >>> Dij[Dij <= 0] <- minDij
> >>>
> >>> Will do this replacement if Dij is your matrix (replace Dij with
> >>> whatever the name of your matrix is). Then supply the new matrix to
> >>> metaMDS. "
> >>>
> >>
> >>
> >> When I use unique function as #x.dist <- dist(unique(X))
> >> this cutted some of my samples that I do not know. I can't see my variables
> >> in the plot. Some help?
> >>
> >> Cheers, Arnaldo.
> >>
> >>
> >> epi<- read.table("epi.txt", h=TRUE,row.names=1)
> >>
> >> library('vegan')
> >> library('MASS')
> >>
> >> epi.<- as.matrix(epi)
> >>
> >> dissim1<- vegdist(epi., method="jaccard", binary=TRUE)
> >> dissim1.mds<-metaMDS(dissim1,k=2)
> >>
> >> plot(dissim1.mds,type="n")
> >> text(dissim1.mds, labels=as.character(row.names(epi)))
> >>
> >> #epi
> >> sample sp1 sp2 sp3 sp4 sp5 sp6 sp7 sp8 sp9
> >> sp10 sp11 sp12 sp13 sp14 sp15 sp16 sp17 sp18 sp19
> >> 1E 1 0 1 0 0 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 2E 0 0 1 0 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 3E 1 0 1 0 0 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 4E 1 0 1 0 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 5E 1 0 0 0 0 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 6E 1 0 0 0 0 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 7M 1 0 1 0 0 0 1 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 8E 1 0 0 0 0 0 1 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 11E 1 0 0 0 1 0 1 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 12E 0 0 0 0 0 1 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 13E 0 0 0 0 0 0 1 0 0 0 1 1 0 0
> >> 0 0 0 0 0
> >> 14E 0 0 1 0 0 0 0 0 0 0 0 1 0 0
> >> 0 0 0 0 0
> >> 15E 1 0 0 0 0 0 1 0 0 0 0 0 1 0
> >> 0 0 0 0 0
> >> 17E 0 0 0 0 1 0 1 0 1 0 0 0 0 0
> >> 0 0 0 0 0
> >> 18E 0 0 0 0 0 0 1 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 20E 1 0 0 0 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 21E 1 0 0 0 0 0 0 0 0 0 0 0 0 1
> >> 0 0 0 0 0
> >> 22E 1 0 0 0 1 0 0 0 1 0 0 0 0 0
> >> 0 0 0 0 0
> >> 23E 1 0 1 1 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 27E 1 0 0 1 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 28E 1 0 0 0 1 0 1 1 0 0 1 0 0 0
> >> 0 0 0 0 0
> >> 30E 1 1 0 0 1 0 0 0 0 0 0 0 0 0
> >> 0 0 0 0 0
> >> 31E 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> >> 1 0 0 0 0
> >> 32E 1 0 0 0 0 0 0 0 1 0 0 0 0 0
> >> 0 0 0 0 0
> >> 9E 0 0 0 0 0 1 1 1 1 1 0 0 0 0
> >> 0 0 1 0 0
> >> 10E 0 0 1 0 0 1 1 0 1 1 1 0 0 0
> >> 0 1 0 1 1
> >>
> >> ------
> >> Arnaldo D`Amaral Pereira Granja Russo
> >> Instituto Ambiental Boto Flipper
> >> www.institutobotoflipper.com.br
> >>
> >> [[alternative HTML version deleted]]
> >>
> >> _______________________________________________
> >> R-sig-ecology mailing list
> >> R-sig-ecology at r-project.org
> >> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>
--
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
Dr. Gavin Simpson [t] +44 (0)20 7679 0522
ECRC, UCL Geography, [f] +44 (0)20 7679 0565
Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk
Gower Street, London [w] http://www.ucl.ac.uk/~ucfagls/
UK. WC1E 6BT. [w] http://www.freshwaters.org.uk
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
More information about the R-sig-ecology
mailing list