[R-sig-ME] error when looping many MCMCglmm runs

Manabu Sakamoto manabu.sakamoto at gmail.com
Sun Nov 30 09:13:55 CET 2014

Dear list,

I'm not quite sure if this is the right list for this, but I only get this
issue when I am running MCMCglmm, so I post it here.

I am wondering if anyone has come across this issue: I am experiencing some
errors when running MCMCglmm in nested loops.

I have three datasets and a set of model formulations for each. So I loop
through the datasets (i) with nested loops (j) for each model formulation,
and I store the results in a list for each dataset. I then save the
workspace, and move onto the second dataset (i+1).

The first time I noticed something strange was after the workspace is saved
at the end of the ith iteration, and I read in the data for the i+1th
dataset. Normally, I'd read in the names of the files using list.files()
and do some regular expressions to select the right filename to read in,
but in this case, the vector of filenames somehow became empty. If I do
this manually, then there's nothing wrong. I thought maybe R was jumping
ahead of itself so at the end of the ith loop I set it so that R sleeps for
1 second, which for some reason resolved this issue.

Today, I noticed something different. I get an error with MCMCglmm stating
that the mixed model was singular. I never get this when I run the same
model on its own or even if inside a loop, if within a single dataset. It
happened when I looped through different datasets. Is there some known
issue about too many loops, and maybe something to do with memory use that
corrupts MCMCglmm runs?

Would setting R to sleep or removing R objects more frequently help resolve
this issue?
I can try these but every loop takes hours so debugging errors that only
occur over multiple nested loops is painstakingly time consuming.

Any suggestions or thoughts would be most appreciated.

Many thanks,

Manabu Sakamoto, PhD
manabu.sakamoto at gmail.com

	[[alternative HTML version deleted]]

More information about the R-sig-mixed-models mailing list