In R: Force `foreach` loop to assign each computation inside the loop to a separate variable-CodePudding

So I have a foreach loop that assigns several different calculations inside the loop to separate variables. While running under a normal for loop, all variables are accessible, but when switching to a foreach loop, the function only returns the last 'stored' variable. How do I change this?

# assume the parallel processing is already set;
# won't affect the result

set.seed(1)

data <- dplyr::tibble(
  x = rnorm(5),
  y = rnorm(5, sd = 0.5),
  z = rnorm(5, sd = 0.25)
)

sims = 10

df = foreach(sim = 1:sims) %dopar% {
  
  calc <- data |> 
          furrr::future_pmap_dfr(
            function(x, y, z) {
              res <- (x   y) / z
              names(res) <- 'col'
              return(res)
            }
          )
  
  calc_2 <- data |>
            furrr::future_pmap_dfr(
              function(x, y, z) {
                res <- (x y z)^2
                names(res) <- 'col'
                return(res)
              }
            )
}

df[[1]] returns:

# A tibble: 5 × 1
    col
  <dbl>
1 0.434
2 0.275
3 0.387
4 1.77 
5 0.210

When switching the orders of calc and calc_2, the function now returns:

# A tibble: 5 × 1
     col
   <dbl>
1 -2.74 
2  4.38 
3  3.00 
4 -3.40 
5  0.629

This loop is not making the other calculation accessible, and when I run the foreach inside my real function -- which operates as intended using a for loop -- the function fails because only one of the three variables assigned in the foreach loop is stored and made accessible throughout the function call. Similar to a for loop, how do I force the foreach loop to assign all variables calculated inside the loop so that they can be carried throughout the function?

CodePudding user response：

You can return the list of results, and process accordingly.

df = foreach(sim = 1:sims) %dopar% {
  
  calc <- data |> 
          furrr::future_pmap_dfr(
            function(x, y, z) {
              res <- (x   y) / z
              names(res) <- 'col'
              return(res)
            }
          )
  
  calc_2 <- data |>
            furrr::future_pmap_dfr(
              function(x, y, z) {
                res <- (x y z)^2
                names(res) <- 'col'
                return(res)
              }
            )
  list(calc, calc_2)
}

df is now a list of length sims, where each element is a list of length 2, containing calc and calc_2