while loop inside tryCatch function

Question

I am running GLMM on permuted data and for some of them I have an error message of convergence. As this is my null model, I just have to resample this specific permuted data. I thus tried to handle the tryCatch function of R but I have some failure. I have Perm which is my permuted data set of data frame and pglmm which is the glmm model for the permuted data. So I have the following structure:

pglmm<-NULL
for (i in seq_along(Perm)){
pglmm[[i]]<-summary(lme("My GLMM model"))
}

To handle the error I tried this structure

pglmm<-NULL
for (i in seq_along(Perm)){
pglmm[[i]]<- tryCatch(summary(lme()),
error=function(err){
pglmm[[i]]<- summary(lme("My GLMM model on resample data"))
return(pglmm[[i]])
}
}

For now it’s working but even by resampling once the data in which I have the error I may have a convergence problem so I need to add a while loop in order to resample until this convergence error doesn’t appear anymore. However I do not really know where to add this while loop (inside error=function(err) or should it be when starting the for loop?).

Thank you in advance for your help.

Answer 1

pre-allocate and fill, increment the iterator on success rather than iterating a sequence

pglmm <- vector("list", length(Perm))
i <- 1
while (i <= length(Perm)) {
    pglmm[[i]] = tryCatch(summary(lme(...)), error=identity)
    if (!is(pglmm[[i]], "error"))
        i <- i + 1
}

or avoid the need to manage memory with

pglmm <- replicate(length(Perm), {
    repeat {
        result <- tryCatch(summary(lme(...)), error=identity)
        if (!is(result, "error"))
            break
    }
    result
})

Answer 2

Maybe something along the following lines

pglmm <- vector("list", length(Perm))
for (i in seq_along(pglmm)){
    while(is.null(pglmm[[i]])) {
        tryCatch(
            pglmm[[i]] <- summary(lme(...)), 
            error = function(e) {print(e); print("retrying...")}
        )
    }
}

pglmm is initialised to a list of same length as Perm, filled with NULL values. Going through each element of the list, you can repeat your data re-sampling and model fitting (this is the pglmm[[i]]<- ... line) until there is no convergence error. Once a successful model run has been achieved, it is assigned to the ith entry of pglmm and the loop then proceeds to the next index, and so on.