Comparing floating point numbers is (at least for me) pretty scary when you get down into it.
Here are some more accessible treatments (although Goldberg’s paper seems to be the definitive treatment on the subject): What Every Programmer Should Know About Floating-Point Arithmetic: http://floating-point-gui.de Comparing Floating Point Numbers, 2012 Edition: https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/ HTH. JWB ________________________________________ Joseph W. Brown Post-doctoral Researcher, Smith Laboratory University of Michigan Department of Ecology & Evolutionary Biology Room 2071, Kraus Natural Sciences Building Ann Arbor MI 48109-1079 josep...@umich.edu > On 16 Aug, 2016, at 12:15, Klaus Schliep <klaus.schl...@gmail.com> wrote: > > We should consider to improve is.ultrametric to use as tolerance a certain > number significant figures (e.g. 8-10 digits). This way tol would only depend > on the tip-to-root distance and not a machine dependent minimal value. I find > the use of significant figures suits really well as a stopping criterion for > example in my ML functions, especially as it works for small and large trees > equally well. > > write.tree has an option to define the number of digits, it will be generally > less than the internal representation and I suspect one could possible export > with higher precision from other programs, but will result in larger file > sizes which one has to put in consideration for MCMC / bootstrap tree files. > > As a side note on the developer site of the r-project.org > <http://r-project.org/> homepage (not CRAN) is a link to this nice ancient > paper (http://www.validlab.com/goldberg/paper.pdf > <http://www.validlab.com/goldberg/paper.pdf>) > > Regards, > Klaus > > > PS: And Martin of course I am happy with Liam's solution as it uses phangorn > ;) > > On Tue, Aug 16, 2016 at 10:22 AM, Joseph W. Brown <josep...@umich.edu > <mailto:josep...@umich.edu>> wrote: > That’s a good tip, but the issue here is that many packages use > is.ultrametric internally with a hard-coded (likely, default) value of tol. > In this case, Martin simply cannot perform the analyses he wants to run > because this option is inaccessible. > > JWB > ________________________________________ > Joseph W. Brown > Post-doctoral Researcher, Smith Laboratory > University of Michigan > Department of Ecology & Evolutionary Biology > Room 2071, Kraus Natural Sciences Building > Ann Arbor MI 48109-1079 > josep...@umich.edu <mailto:josep...@umich.edu> > > > >> On 16 Aug, 2016, at 10:15, Klaus Schliep <klaus.schl...@gmail.com >> <mailto:klaus.schl...@gmail.com>> wrote: >> >> Hello all, >> this may come be surprising to many, but consulting the manual >> ?is.ultrametric can be helpful. Why not simply try e.g. is.ultrametric(tree, >> tol=.01) ???? >> So in this sense RTFM >> Regards, >> Klaus >> >> >> On Aug 16, 2016 9:31 AM, "Martin Dohrmann" <m.dohrm...@lrz.uni-muenchen.de >> <mailto:m.dohrm...@lrz.uni-muenchen.de>> wrote: >> >> Am 16.08.2016 um 15:20 schrieb Joseph W. Brown: >> >> > I agree that it is almost certainly numerical precision, as rescaling the >> > tree fixes things: >> > >> > > is.ultrametric(phy); >> > [1] FALSE >> > > fie <- phy; >> > > fie$edge.length <- fie$edge.length * 0.1; >> > > is.ultrametric(fie); >> > [1] TRUE >> > >> > I’ve also tested the ultrametricity(?) with a non-R program and the >> > results are the same. So can we assume this is not a PhyloBayes issue, and >> > rather an unavoidable problem with large, old trees? But the fact that >> > MrBayes trees are ultrametric is confusing; does MrBayes use less precise >> > edge lengths? >> >> Actually the MrBayes branch lengths appear to be MORE precise. >> >> Martin >> >> >> >> >> On 16 Aug, 2016, at 08:53, Liam J. Revell <liam.rev...@umb.edu >> >> <mailto:liam.rev...@umb.edu>> wrote: >> >> >> >> Hi Martin. >> >> >> >> Since you are writing & reading trees to file, my guess is that it has to >> >> do with numerical precision - that is, the rounding of your edge lengths >> >> when they are written to file. >> >> >> >> Does your tree look ultrametric when plotted in R? If so, this is >> >> probably the case. >> >> >> >> My recommendation is that you use phangorn to compute the non-negative >> >> least-squares edge lengths with the condition that the tree is >> >> ultrametric. This will give you the edge lengths that result in the >> >> distances between taxa with minimum sum of squared differences from the >> >> distances implied by your input tree, under the criterion that the >> >> resulting tree is ultrametric. >> >> >> >> To do this you need to merely run: >> >> >> >> library(phytools) >> >> library(phangorn) >> >> is.ultrametric(tree) ## fails >> >> plotTree(tree,ftype="off") ## does my tree look ultrametric? >> >> nnls<-nnls.tree(cophenetic(tree),tree,rooted=TRUE) >> >> is.ultrametric(tree) ## should pass >> >> >> >> Let us know if this works. All the best, Liam >> >> >> >> Liam J. Revell, Associate Professor of Biology >> >> University of Massachusetts Boston >> >> web: http://faculty.umb.edu/liam.revell/ >> >> <http://faculty.umb.edu/liam.revell/> >> >> email: liam.rev...@umb.edu <mailto:liam.rev...@umb.edu> >> >> blog: http://blog.phytools.org <http://blog.phytools.org/> >> >> >> >> On 8/16/2016 6:41 AM, Martin Dohrmann wrote: >> >>> Hi, >> >>> I want to do some diversification pattern analyses with various R >> >>> packages. I'm using a time-calibrated tree produced by PhyloBayes. >> >>> Branch lengths are in millions of years and all taxa are extant, so the >> >>> tree should be ultrametric. However, when I call "is.ultrametric", it >> >>> returns "FALSE". >> >>> >> >>> Has anybody encountered something like this? Any ideas about what's >> >>> going on/how to solve this? >> >>> >> >>> Some further information: >> >>> I also tried other PhyloBayes time trees, with the same result. In >> >>> contrast, MrBayes time trees I tried for comparison are recognized as >> >>> ultrametric. Regarding my diversification analyses, TESS would not run, >> >>> telling me "The likelihood function is only defined for ultrametric >> >>> trees!". On the other hand, BAMMtools doesn't seem to have a problem >> >>> with my tree. I haven't tried other packages yet, but I suspect RPANDA, >> >>> TreePar etc. might also have issues if they don't recognize my tree as >> >>> ultrametric. >> >>> >> >>> I'd appreciate any help! >> >>> >> >>> Best wishes, >> >>> Martin >> >>> >> >>> Dr. Martin Dohrmann >> >>> Ludwig-Maximilians-University Munich >> >>> Dept. of Earth & Environmental Sciences >> >>> Palaeontology & Geobiology >> >>> Molecular Geo- & Palaeobiology Lab >> >>> Richard-Wagner-Str. 10 >> >>> 80333 Munich, Germany >> >>> Phone: +49-(0)89-2180-6593 <tel:%2B49-%280%2989-2180-6593> >> >>> >> >>> >> >>> [[alternative HTML version deleted]] >> >>> >> >>> _______________________________________________ >> >>> R-sig-phylo mailing list - R-sig-phylo@r-project.org >> >>> <mailto:R-sig-phylo@r-project.org> >> >>> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> >>> <https://stat.ethz.ch/mailman/listinfo/r-sig-phylo> >> >>> Searchable archive at >> >>> http://www.mail-archive.com/r-sig-phylo@r-project.org/ >> >>> <http://www.mail-archive.com/r-sig-phylo@r-project.org/> >> >>> >> >> >> >> _______________________________________________ >> >> R-sig-phylo mailing list - R-sig-phylo@r-project.org >> >> <mailto:R-sig-phylo@r-project.org> >> >> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> >> <https://stat.ethz.ch/mailman/listinfo/r-sig-phylo> >> >> Searchable archive at >> >> http://www.mail-archive.com/r-sig-phylo@r-project.org/ >> >> <http://www.mail-archive.com/r-sig-phylo@r-project.org/> >> > >> >> Dr. Martin Dohrmann >> Ludwig-Maximilians-University Munich >> Dept. of Earth & Environmental Sciences >> Palaeontology & Geobiology >> Molecular Geo- & Palaeobiology Lab >> Richard-Wagner-Str. 10 >> 80333 Munich, Germany >> Phone: +49-(0)89-2180-6593 <tel:%2B49-%280%2989-2180-6593> >> >> >> [[alternative HTML version deleted]] >> >> >> _______________________________________________ >> R-sig-phylo mailing list - R-sig-phylo@r-project.org >> <mailto:R-sig-phylo@r-project.org> >> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> <https://stat.ethz.ch/mailman/listinfo/r-sig-phylo> >> Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ >> <http://www.mail-archive.com/r-sig-phylo@r-project.org/> > > > > > -- > Klaus Schliep > Postdoctoral Fellow > Revell Lab, University of Massachusetts Boston > http://www.phangorn.org/ <http://www.phangorn.org/> > [[alternative HTML version deleted]] _______________________________________________ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
