Question

I would like to transform the following nested for loop

first <- c(1, 2, 3)
second <- c(1, 2, 3)

dummy = matrix(double(), len(first), len(second))
c <- list()
c$sum <- dummy
c$times <- dummy

for (i in 1:len(first)) {
    for (j in 1:len(second)) {
        c$sum[i, j] <- first[i] + second[j]
        c$times[i, j] <- first[i] * second[j]
    }
}

c

into code using foreach and get the same list of matrices as a result. I tried many different things but the closest “result” is this:

x <- foreach(b = second, .combine = "cbind") %:% foreach(a = first, .combine = "c") %do% {
            c <- list()
            c$sum <- a+b
            c$times <- a*b
            out <- c
            }
x

How to get this list of matrices right using foreach?

EDIT: One possibility is using a result and transform it after calling foreach:

res <- list()
res$sum <- x[rownames(x)=="sum", ]
rownames(res$sum) <- NULL
colnames(res$sum) <- NULL
res$times <- x[rownames(x)=="times", ]
rownames(res$times) <- NULL
colnames(res$times) <- NULL
res

How to “parametrize” foreach so there is no need to transform results?

Answer 1

You “just” have to provide the correct .combine function.
If you only have numbers, you can return an array rather than a list.

library(foreach)
library(abind)
first <- 1:3
second <- 4:5
x <- 
  foreach(b = second, .combine = function(...) abind(..., along=3)) %:% 
  foreach(a = first,  .combine = rbind) %do% {
    c( sum=a+b, times=a*b )
  }

If you really need lists, writing the combining functions is much harder.
Instead, you can build a data.frame, and reshape it afterwards, if needed.

x <- 
  foreach(b = second, .combine = rbind) %:% 
  foreach(a = first,  .combine = rbind) %do% {
    data.frame(a=a, b=b, sum=a+b, times=a*b )
  }
library(reshape2)
list(
  sum   = dcast(x, a ~ b, value.var="sum"  )[,-1],
  times = dcast(x, a ~ b, value.var="times")[,-1]
)