You're at best on the border of how many parameters you can estimate given the size of your dataset. One thing you might try is running OUwie, which also implements that model but uses a different optimizer and starting values, though I wouldn't be surprised if you get similar answers. Trying a grid of values for a contour plot can tell you something about the likelihood surface you're trying to optimize on; it could be very flat. We had this function in OUwie until a license conflict with akima made us have to cut this functionality, but it's easy enough to do.
Best, Brian _______________________________________ Brian O'Meara Assistant Professor Dept. of Ecology & Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Sat, Jun 29, 2013 at 9:03 AM, Pascal Title <pti...@umich.edu> wrote: > Hello, > > Given a phylogeny with 29 tips, and 5 traits that each fit a OU model > better than a BM model of trait evolution (according to fitContinuous in > GEIGER), I would like to fit a multivariate OU model to these 5 traits > using the hansen() function in the OUCH package in R. The goal is to get > multivariate rate values. I am using R v3.0.1 and version 2.8-2 of the OUCH > package. > > Under default settings, I get an "unsuccessful convergence" error. > By changing various settings, I stop getting a convergence error message. > But the alpha and sigma2 values seem to be unrealistically large: > > *hansen(data = otd[colnames(data)], tree = ot, regimes = otd["regimes"], * > * sqrt.alpha = priors, sigma = priors, method = "L-BFGS-B", * > * maxit = 5000)* > * > * > * optim diagnostic message: CONVERGENCE: REL_REDUCTION_OF_F <= > FACTR*EPSMCH* > *alpha:* > * [,1] [,2] [,3] [,4] [,5]* > *[1,] 61.385945 17.5563426 -33.7110759 3.697183 50.622310* > *[2,] 17.556343 45.9034684 -0.3705461 27.995283 -6.633346* > *[3,] -33.711076 -0.3705461 42.8040546 8.072352 -40.675454* > *[4,] 3.697183 27.9952829 8.0723524 25.435097 -12.115062* > *[5,] 50.622310 -6.6333460 -40.6754540 -12.115062 59.536768* > * > * > *sigma squared:* > * [,1] [,2] [,3] [,4] [,5]* > *[1,] 89.262795 32.572924 9.099638 28.769264 24.711740* > *[2,] 32.572924 40.506308 13.271956 19.279625 -9.633395* > *[3,] 9.099638 13.271956 4.387822 5.986198 -3.965904* > *[4,] 28.769264 19.279625 5.986198 11.966763 2.241925* > *[5,] 24.711740 -9.633395 -3.965904 2.241925 18.995603* > > I've been told that sigma2/(2*alpha) should be approximately equal to the > variance of the trait. If I try this for the first trait, sigma2/(2*alpha) > = 0.727, and the actual variance of the first trait is 1.304. So that > doesn't match, and values for alpha and sigma2 are extremely large. > > My question is, where do I go from here? What can I do to get more > realistic values? I've heard that you can give different starting values to > the sqrt.alpha and sigma arguments of the hansen function, but here these > are lower triangular matrices, so I don't know what to adjust, or what > other starting values to give, or how to know when I've found the most > realistic values I'm going to find. > > >From the following link, you can download the tree, data and script to see > for yourself: > https://dl.dropboxusercontent.com/u/34644229/Pascal_OUCH.zip > > Any advice or suggestions would be greatly appreciated! > > Thank you, > > -Pascal > > > > > -- > Pascal Title, MSc. > PhD student, Rabosky Lab < > http://www-personal.umich.edu/~drabosky/Home.html> > Dept of Ecology and Evolutionary Biology > University of Michigan > pti...@umich.edu > > [[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/ > [[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/
