How to reshape output of simulate function using purrr-CodePudding

I would like to generate a wide data frame of simulated observations using the stats::simulate function and the purrr package. Here's the basic idea:

library(tidyverse)

mod <- lm(mpg ~ cyl   disp, data = mtcars)

m_sim <- function(s) {
  stats::simulate(mod, nsim = 1, newdata = mtcars, seed = s) 
}

df <- map_dfr(1:3, m_sim)

The problem is that this code generates s rows for each observation in the data set. For example, where s = 3 (as above):

df %>% 
  rownames_to_column(var = "car") %>% 
  dplyr::filter(str_sub(car, 1, 10) == "Datsun 710")
#>               car    sim_1
#> 1  Datsun 710...3 23.53562
#> 2 Datsun 710...35 30.94046
#> 3 Datsun 710...67 26.87957

How can I fix the code so that the simulations appear as columns? Here, the desired output would be:

#>          car     sim_1     sim_2     sim_3
#> 1 Datsun 710  23.53562  30.94046  26.87957

CodePudding user response：

Instead of _dfr, use _dfc or the more recent list_cbind

library(purrr)
library(dplyr)
library(stringr)
map(1:3, m_sim) %>% 
  list_cbind%>%
   rownames_to_column(var = "car") %>%
   dplyr::filter(str_sub(car, 1, 10) == "Datsun 710")

CodePudding user response：

You can add a line to the function to rename the column using the seed, then use map_dfc() to column-bind:

library(dplyr)
library(purrr)
library(stringr)

m_sim <- function(s) {
  stats::simulate(mod, nsim = 1, newdata = mtcars, seed = s) %>% 
    rename("sim_{s}" := sim_1)
}

map_dfc(1:3, m_sim) %>% 
  rownames_to_column(var = "car") %>% 
  dplyr::filter(str_sub(car, 1, 10) == "Datsun 710")

         car    sim_1    sim_2    sim_3
1 Datsun 710 23.53562 30.94046 26.87957