Specifying starting point for MCMCglmm()

Dear Sree~

Yes, there are 200 rows of data in the dataset. We are assuming a gaussian
distribution in our models. If we were taking a frequentist approach, it
would be a glmm (mixed model).



Thank you,

*From: *sree datta <sreedta8 at gmail.com>
*Date: *Friday, June 5, 2020 at 4:46 PM
*To: *"Ruhs, Emily" <ecruhs at usf.edu>
*Cc: *"r-sig-mixed-models at r-project.org" <r-sig-mixed-models at r-project.org

*Subject: *Re: [R-sig-ME] Specifying starting point for MCMCglmm()



This email originated from outside of USF. Do not click links or open
attachments unless you recognize the sender or understand the content is
safe.



When you say 200 records, is this the total number of rows in the dataset?
What distribution are you assuming for your dependent variables? What
frequentist approach would you use instead of MCMC to model this data?



Sree



On Fri, Jun 5, 2020 at 3:59 PM Ruhs, Emily <ecruhs at usf.edu> wrote:

Hi Sree~

Thank you for reaching out.



I am running 5 models that are near the form:

EC1<-MCMCglmm(cbind(Infl.X, Infl.Y, Prop.Bottom, Prop.Top, Log.Slope,
Coef) ~ (trait):log10Br+(trait):log10Bo,

              random = ~us(trait):phylo, family=rep("gaussian", 6),
rcov=~us(trait):units,

              ginverse=list(phylo=inv.phyloEC$Ainv), prior=prior.1,
data=speciesEC,

              nitt=NITT*Mult,thin=THIN*Mult,burnin=BURN*Mult)



I have 6 response variables and our models cover the full sweet of
possibilities with log10Br and log10Bo as predictor variables (i.e. null
model, log10Br alone, log10Bo alone, log10Br+log10Bo, log10Br*log10Bo). I
have about 200 records (species) in the dataset.



I?ve gotten results from the models using
Mult=7;NITT=260000;THIN=200;BURN=60000, but when we plot the trace and
density, they still do not appear to be converging. Therefore I would like
to extend to a Mult = 10.



Therefore, I was hoping to use the start= command to extend the models,
without having to restart them; however, maybe that isn?t possible. Any
advice you can provide would be greatly appreciated!

Best,

*From: *sree datta <sreedta8 at gmail.com>
*Date: *Friday, June 5, 2020 at 3:48 PM
*To: *"Ruhs, Emily" <ecruhs at usf.edu>
*Cc: *"r-sig-mixed-models at r-project.org" <r-sig-mixed-models at r-project.org

*Subject: *Re: [R-sig-ME] Specifying starting point for MCMCglmm()



This email originated from outside of USF. Do not click links or open
attachments unless you recognize the sender or understand the content is
safe.



Hi Emily



What are the specifications of your data you are using in the model that
it is taking so long (14 days)? How many variables and records are you
using? Have you attempted to run the model with a smaller subset of the
data? If yes, what were the results?



Sree





On Thu, Jun 4, 2020 at 4:10 PM Ruhs, Emily <ecruhs at usf.edu> wrote:

Hi everyone~

I am running a series of large models in R using MCMCglmm. After running
the models for 14 days (complicated, multivariate models), I found that the
models have not converged yet and I need to run more iterations. I know you
can use the start= to specify a starting function, but I?m having
difficulties getting the model to run.

model_EC <- parLapply(cl=cl,listEC, function(i) {
  MCMCglmm(i[[1]],
           random = ~us(trait):phylo, family=rep("gaussian", 6),
rcov=~us(trait):units,

 ginverse=list(phylo=inv.phyloEC$Ainv),prior=prior.1,data=i[[2]],
start=1820000,
           nitt=2600000,thin=1400,burnin=0)})

As the code is written, I have the start=1820000, which is where the last
iteration left off. Can anyone explain what I?m doing wrong in specifying
the start function?

Any help or suggestions would be greatly appreciated!

Best,
Emily Cornelius Ruhs
Postdoctoral Scholar
University of South Florida



        [[alternative HTML version deleted]]