[R] Referencing objects within a loop

[Douglas Bates]

Doran, Harold wrote:
> Dear List:
> 
> It appears that simulating data where all dataframes are stored as a
> list will only work for relatively small analyses. Instead, it appears
> that creating N individual dataframes, saving them, and loading them
> when needed is the best way to save memory and make this a feasible
> task.
> 
> As such, I now have a new(er) question with respect to dealing with
> individual files within a loop. To begin, I construct 250 individual
> data files as follows:
> 
> library(MASS)
> Sigma<-matrix(c(400,80,80,80,80,400,80,80,80,80,400,80,80,80,80,400),4,4
> ) 
> mu<-c(100,150,200,250) 
> sample.size<-5000
> N=250 #Number of dataframes 
> 
> # Step 1 Create dataframes
> for(i in 1:N) 
> { 
> assign(paste("Data.", i, sep=''), 
> as.data.frame(cbind(seq(1:sample.size),(mvrnorm(n=sample.size, mu,
> Sigma))))) 
> } 
> 
> 
> My goal is to now save each file, remove them from memory, and load each
> file individually, run it through a linear model, and save the output.
> 
> To attempt to save each file I try
> 
> for (i in 1:N) {
> save(paste("Data.", i, sep=""), file=paste("Data.",i, ".Rdata", sep=""))
> 
> }
> 
> Which gives me an error message. It essentially states that Object
> "paste("Data.", i, sep = "")" not found
> 
> I'm not quite sure what I am doing wrong here. 

You need to use 'get' to convert a character string into an object.  An 
alternative is

for (nm in paste("Data.", 1:N, sep = '')) {
   save(list = nm, file = paste(nm, ".Rdata", sep = ''))
}

> The other issue I am encountering is how to best work with an object in
> a loop. I get the following to work, but I'm not sure if this is the
> best method for doing so. 
> 
> for(i in 1:N){
> assign(paste("out.",i,sep=""),
> lm((get(paste("Data.",i,sep=""))[["V3"]])~(get(paste("Data.",i,sep=""))[
> ["V2"]])))
> }

As shown above, looping in R does not need to be over a numeric index 
vector.  You can loop over any vector object, including the vector of 
names.  Another way you can simplify this is to use the formula/data 
specification for lm.  If your model formula is always going to be V3 ~ 
V2 then just use that as in

for (nm in paste("Data.", 1:N, sep = '')) {
   assign(paste(nm, '.out', sep=''), lm(V3 ~ V2, data = get(nm))
}

There are probably better approaches but I think you would prefer that I 
work on the lme4 code instead of spending more time answering these 
questions.  :-)

(BTW, I believe I have a very general formulation for fitting models 
with non-nested grouping factors, such as the student-teacher-school 
data from DC.  Still a few bugs though.)