[R-sig-ME] Specifying starting point for MCMCglmm()

[Pierre de Villemereuil]

Hi,

The "start" argument of MCMCglmm() cannot be used to restart at a given iteration (and if it did, it would require a saved model to do so, since the algorithm is not deterministic).

>From the function help and "start" section:

> optional list having 4 possible elements: ‘R’ (R-structure)
> ‘G’ (G-structure) and ‘liab’ (latent variables or
> liabilities) should contain the starting values where ‘G’
> itself is also a list with as many elements as random effect
> components. The fourth element ‘QUASI’ should be logical: if
> ‘TRUE’ starting latent variables are obtained heuristically,
> if ‘FALSE’ then they are sampled from a Z-distribution

So, you would need to extract such components from an already saved model. R and G components would be located in "model$VCV". Unless I'm misguided, you need to set "pl = TRUE" to save the latent variables that the "liab" component needs, which are then saved in the "model$Liab" component.

Note that I don't believe the information you can provide to "start" will be enough to restore your chain *exactly* at the state it was, but it should place you near enough a convergent state if drawn from a convergent state.

Cheers,
Pierre.


Le jeudi 4 juin 2020, 22:05:46 CEST Ruhs, Emily a écrit :
> 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]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>