CODEWITHSUNDEEP
rtidyversepurrr
Demetri Pananos
Demetri Pananos

Reputation: 7394

Using pmap with specific columns of a dataframe

I want to simulate some coin flips of a fair and unfair coin using purrr. I have parameters for rbinom as columns in a dataframe. Here is some code I've started with:

library(tidyverse)

#setting up params for rbinom
params = list(size = 600, n = 1, p_fair = 0.5,p_unfair = 0.6)


params %>% 
  purrr::map_df(~rep(., times = 10))

How can I use purrr to pass different parameters p_fair/p_unfair into rbinom and put the results as a column in my dataframe? I would like the result to look like:

 size     n p_fair p_unfair  fair unfair
   <dbl> <dbl>  <dbl>    <dbl> <int>  <int>
 1   600     1    0.5      0.6   308    373
 2   600     1    0.5      0.6   305    367
 3   600     1    0.5      0.6   280    367
 4   600     1    0.5      0.6   299    374
 5   600     1    0.5      0.6   298    360
 6   600     1    0.5      0.6   298    346
 7   600     1    0.5      0.6   301    359
 8   600     1    0.5      0.6   292    376
 9   600     1    0.5      0.6   300    347
10   600     1    0.5      0.6   305    357

Upvotes: 1

Views: 1300

Answers (2)

Nettle
Nettle

Reputation: 3321

The straightforward approach isn't as elegant as some other answers, but more readable (and not much longer). 

# Libraries
library(tidyverse)

# Data
params = list(size = 600, n = 1, p_fair = 0.5,p_unfair = 0.6)
df1 <- params %>% map_df(~rep(., times = 10))

Here's a reminder of the function and it's arguments

# rbinom(n, size, prob)

Create fair/unfair dataframes that contain all arguments (each column header named for the relevant argument).

p_fair_vars  <- df1 %>% 
  select(n, size, p = p_fair)

p_unfair_vars <- df1 %>% 
  select(n, size, p = p_unfair)

Apply dataframes using pmap.

df1 %>% 
  mutate(fair = pmap_int(p_fair_vars, rbinom),
         unfair = pmap_int(p_unfair_vars, rbinom) )

#> # A tibble: 10 x 6
#>     size     n p_fair p_unfair  fair unfair
#>    <dbl> <dbl>  <dbl>    <dbl> <int>  <int>
#>  1   600     1    0.5      0.6   296    351
#>  2   600     1    0.5      0.6   311    351
#>  3   600     1    0.5      0.6   290    356
#>  4   600     1    0.5      0.6   294    359
#>  5   600     1    0.5      0.6   294    348
#>  6   600     1    0.5      0.6   306    365
#>  7   600     1    0.5      0.6   288    370
#>  8   600     1    0.5      0.6   305    366
#>  9   600     1    0.5      0.6   314    327
#> 10   600     1    0.5      0.6   303    356

...or alternately:

df1 %>% 
  mutate(fair_vars = str_c(size, n, p_fair, sep="," ),
         unfair_vars = str_c(size, n,   p_unfair, sep = ","),
         fair = pmap_int(p_fair_vars, rbinom),
         unfair = pmap_int(p_unfair_vars, rbinom)) %>% 
  select(-c(5:6) )

Created on 2018-11-03 by the reprex package (v0.2.1)

Upvotes: 0

akrun
akrun

Reputation: 886938

We can loop through the 'p_' columns and apply the rbinom

library(dplyr)
library(purrr)
df1 %>%
    select(matches('p_')) %>%
    map(~ df1 %>%
            select(n, size) %>% 
            mutate(p = .x) %>%
            pmap_int(rbinom)) %>% 
    bind_cols %>%
    rename_all(funs(sub("p_", "", .))) %>%
    bind_cols(df1, .)

Another option would be to convert to 'long' format with gather, apply the rbinom and then reshape it back to 'wide'

library(tidyr)
df1 %>%
  gather(key, p, p_fair:p_unfair) %>%
  mutate(rval = pmap_int(.[c('n', 'size', 'p')], rbinom)) %>% 
  separate(key, into = c('key1', 'key2')) %>%
  select(-key1) %>% 
  group_by(key2) %>% 
  mutate(n1 = row_number()) %>%
  select(-p) %>% 
  spread(key2, rval) %>% 
  select(-n1, -n, -size) %>%
  bind_cols(df1, .)

data

df1 <- params %>% 
            map_df(~rep(., times = 10))

Upvotes: 1

Related Questions