Passing prior parameters as variables instead of lists

[brianhuey]

First off, thanks so much for writing this package - it's become indispensable in my work.

Both blmer() and bglmer() allow a list to be passed to cov.prior, i.e. bglmer(...,cov.prior=list(fc.nm ~ dist1, fc.nm ~ dist2)), but if we were to set the list as a variable:

cov.prior <-  list(fc.nm ~ dist1, fc.nm ~ dist2)
bglmer(...,cov.prior=cov.prior)

An error is thrown. Some cursory debugging indicates that when a variable that references a list is passed, it gets wrapped in a list twice, which causes problems in evaluateCovPriors(). I was able to hack together a solution but I wanted to raise the issue since there might need to be a more general solution.

[vdorie]

Thanks for bringing this to my attention. I've pushed a commit (ed7f51ba9911be8aa8575eead7833e02c5f043df) that I hope will fix the issue. Let me know if you encounter any problems.

[brianhuey]

It's working for me, thanks so much!

[randel]

Hi Vincent, thanks for the nice package! blmer() works if running in a single instance, but I still got an error if running blmer() in parallel like in the environment of foreach(). Do you have any idea how to fix this? Thanks!

library(blme)
data("sleepstudy", package = "lme4")

scale0 = diag(0.5, 2)

fm4 <- blmer(Reaction ~ Days + (1 + Days|Subject), sleepstudy,
              cov.prior = invwishart(df = 5, scale = scale0))

library(doParallel)
cl <- makeCluster(1)
registerDoParallel(cl)

res = foreach(i = 1) %dopar% {library(blme); blmer(Reaction ~ Days + (1 + Days|Subject), sleepstudy,
                                   cov.prior = invwishart(df = 5, scale = scale0))}

Error in { : task 1 failed - "object 'scale0' not found"

[randel]

It seems that the error can be somehow avoided using the snow package for parallel computing.

library(snow)
clus <- makeCluster(1)

clusterExport(clus, "scale0")
aa <- parSapply(clus, 1, function(x) {
  library(blme); blmer(Reaction ~ Days + (1 + Days|Subject), sleepstudy,
                       cov.prior = invwishart(df = 5, scale = scale0))
})

clusterExport() loads the variable to .GlobalEnv by default, so this may be still an enviromental issue.

[vdorie]

Yeah, I think it has to do with variable scoping. It works with doParallel and foreach with

res <- foreach(i = 1) %dopar% {
  library(blme)
  blmer(Reaction ~ Days + (1 + Days|Subject), sleepstudy,
        cov.prior = invwishart(df = 5, scale = diag(0.5, 2)))
}

[randel]

Yes, the issue is that an "outside" variable cannot be passed to the prior. clusterExport() also cannot work when each run needs its own prior that is saved as variables.