R help - Jun 2013 - Optim seems not to work properly in foreach

Hi guys,

I am working now for several years in R and I would say I manage things pretty
easily, but with the foreach loop I have my problems. I call for a simulation a
double foreach loop and this works fine. Inside the second loop (which I plan to
parallelize later on) I call Rs optim-function:

"simulateML" <- function(sim.it = 250, input.path,
                        output.path, T = 390, ncore = 2) {

        ## load packages ##
        library("mmstruct")
        library("doMC")
        library("foreach")

        ## read input parameters ##
        input <- read.csv(input.path, header = FALSE)

        ## create container to store results ##
        results <- data.frame(NA, nrow = NROW(input) * sim.it, ncol = 12)

        ## initialize the parallel backend ##
        registerDoMC(ncore)

        result.list <- foreach(i = 1:2) %do% {
                input <- as.matrix(input)
                input.row <- input[i, ]
                list <- foreach(i = 1:2, .combine = rbind, .packages =
c("mmstruct", "stats")) %do% {
                        data <- simulateEKOP(size = input.row[1], alpha =
input.row[2],
                                        epsilon = input.row[1], delta =
input.row[4],
                                        mu = input.row[5], T = T)
                        data <- data[,4]
                        ## create start values ##
                        tmp <- mean(data)/T
                        startpar <- c(0, tmp * 0.75/2, tmp * 0.25/2)

                        ## set options for optimization ##
                        optim_fn        <- computeKokotLik
                        optim_method    <- "L-BFGS-B"
                        optim_lower     <- c(-1e+7, 0, 0)
                        optim_upper     <- rep(1e+7, 3)
                        optim_fnscale   <- -1
                        optim_maxit     <- 200
                        optim_ctrl      <- list(fnscale = optim_fnscale,
maxit = optim_maxit)

                        ## start optimization ##
                        res <- optim(par = startpar, fn = optim_fn, data =
data, T = T,
                                        methodLik = "approx", method =
optim_method,
                                        lower = optim_lower, upper =
optim_upper,
                                        control = optim_ctrl, hessian = TRUE)
                        res$par

                }
                print(list)
        }

}

Data simulation and thecreation of startpar works fine, but the parameters in
res$par are always the start parameters. If I run the same commands directly on
the shell I get in res$par the optimized parameters - only inside the foreach
loop optim seems not to work. What could that be?


Best 

Simon

On Mon, Jun 3, 2013 at 11:37 AM, Simon Zehnder <szehnder@uni-bonn.de>

... [Some not minimal, self contained, reproducible code]...

> Data simulation and thecreation of startpar works fine, but the parameters
> in res$par are always the start parameters. If I run the same commands
> directly on the shell I get in res$par the optimized parameters - only
> inside the foreach loop optim seems not to work. What could that be?
>
> Don't know, but but this makes me doubt it has anything to do with
optim
being inside foreach:

fr <- function(x) {
         x1 <- x[1] ; x2 <- x[2]
         100 * (x2 - x1 * x1)^2 + (1 - x1)^2
     }
grr <- function(x) {
         x1 <- x[1] ; x2 <- x[2]
         c(-400 * x1 * (x2 - x1 * x1) - 2 * (1 - x1) , 200 * (x2 - x1 * x1))
     }
library("doMC")
registerDoMC(2)
RNGkind("L'Ecuyer")
set.seed(54321)
foreach(i = 1:2) %do% {
  ret <- foreach(j = 1:2) %do%{
           strtpar <- c(-2,2)+rnorm(2)
           optim(strtpar, fr, grr, method
"L-BFGS-B",control=list(trace=TRUE))$par
          }
  ret
}

Also, wouldn't you want to register 4 cores by default if nesting 2 loops
of 2 ? (to comment on the wisdom of doing so in terms of overhead is beyond
my expertise)

HTH


> Best
>
> Simon
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
	[[alternative HTML version deleted]]