在 mutate 语句中动态引用列名 - dplyr

Question

I stat apologize for the long question, but after quite a while I couldn't figure out a solution myself.

I have this toy dataframe

set.seed(23)
df <- tibble::tibble(
  id = paste0("00", 1:6),
  cond = c(1, 1, 2, 2, 3, 3),
  A_1 = sample(0:9, 6, replace = TRUE), A_2 = sample(0:9, 6, replace = TRUE), A_3 = sample(0:9, 6, replace = TRUE),
  B_1 = sample(0:9, 6, replace = TRUE), B_2 = sample(0:9, 6, replace = TRUE), B_3 = sample(0:9, 6, replace = TRUE),
  C_1 = sample(0:9, 6, replace = TRUE), C_2 = sample(0:9, 6, replace = TRUE), C_3 = sample(0:9, 6, replace = TRUE)
)

# A tibble: 6 x 11
#   id     cond   A_1   A_2   A_3   B_1   B_2   B_3   C_1   C_2   C_3
#   <chr> <dbl> <int> <int> <int> <int> <int> <int> <int> <int> <int>
# 1 001       1     6     3     9     5     0     5     6     0     6
# 2 002       1     4     5     0     8     5     0     1     6     6
# 3 003       2     4     2     8     8     8     6     5     2     5
# 4 004       2     4     4     0     7     2     6     7     5     7
# 5 005       3     1     7     0     9     9     0     5     7     8
# 6 006       3     3     8     7     0     2     5     0     9     4

I would like to create three variables A_def , B_def , C_def that take the values of only one of the corresponding variables < LETTER_NUMBER > depending on the condition that their suffix is equal to variable cond .

For instance, for the rows where cond == 1 , A_def should have values from A_1 , B_def should have values from B_1 , C_def should have values from C_1 . Likewise, if cond == 2 , the *_def columns should have values from the respective *_2 variables.

I managed to achieve my desired output in two ways: one hard-coded (possibly to avoid if cond contains many values) and one using tidyr 's pivoting functions.

Hard-coded solution:

df %>% 
  mutate(
    A_def = ifelse(cond == 1, A_1, ifelse(cond == 2, A_2, A_3)),
    B_def = ifelse(cond == 1, B_1, ifelse(cond == 2, B_2, B_3)),
    C_def = ifelse(cond == 1, C_1, ifelse(cond == 2, C_2, C_3))
  ) %>% 
  select(id, cond, contains("_def"))

tidyr 's solution:

df %>% 
  pivot_longer(cols = contains("_")) %>% 
  mutate(
    number = gsub("[A-Za-z_]", "", name),
    name = gsub("[^A-Za-z]", "", name)
  ) %>% 
  filter(cond == number) %>% 
  pivot_wider(id_cols = c(id, cond), names_from = name, values_from = value, names_glue = "{name}_def")

Output in both cases

# A tibble: 6 x 5
#   id     cond A_def B_def C_def
#   <chr> <dbl> <int> <int> <int>
# 1 001       1     6     5     6
# 2 002       1     4     8     1
# 3 003       2     2     8     2
# 4 004       2     4     2     5
# 5 005       3     0     0     8
# 6 006       3     7     5     4

---

Now, I was wondering whether it is possible to obtain the same output using mutate and/or across in a dynamic fashion (maybe using ifelse statements inside mutate ?). I tried the following code snippets but the results were not as expected. In one of them I tried to make the variable names as symbols within ifelse statements, but I got an error.

df %>% 
  mutate(across(paste0(c("A", "B", "C"), "_1"),
                ~ifelse(cond == 1, cur_column(), 
                        ifelse(cond == 2, cur_column(), paste0(gsub("[^A-Za-z]", "", cur_column()), "_3"))))) %>% 
  select(id, cond, contains("_1"))

df %>% 
  mutate_at(paste0(c("A", "B", "C"), "_1"),
            ~ifelse(cond == 1, ., ifelse(cond == 2, ., paste0(., "_2")))) %>% 
  select(id, cond, contains("_1"))

df %>% 
  mutate_at(paste0(c("A", "B", "C"), "_1"),
            ~ifelse(cond == 1, !!!rlang::syms(paste0(c("A", "B", "C"), "_1")),
                    ifelse(cond == 2, !!!rlang::syms(paste0(c("A", "B", "C"), "_2")),
                           !!!rlang::syms(paste0(c("A", "B", "C"), "_3")))))

Question: is there a way to obtain the same desired output as above using dplyr 's statements such as mutate (or its superseded scoped variants) and/or across ?

Answer 1

As Ronak said, your tidyr solution seems quite fine.

You can simplify it a bit though:

df %>% 
  pivot_longer(cols = contains("_"), names_to = c("name", "number"), names_sep = "_") %>% 
  filter(cond == number) %>% 
  pivot_wider(id_cols = c(id, cond), names_glue = "{name}_def")


## A tibble: 6 x 5
#  id     cond A_def B_def C_def
#  <chr> <dbl> <int> <int> <int>
#1 001       1     7     8     1
#2 002       1     2     5     2
#3 003       2     4     2     3
#4 004       2     0     3     1
#5 005       3     9     0     7
#6 006       3     9     7     0

Answer 2

I agree with the other comments that tidyr makes for more readable code, but here's an alternative approach with pmap :

library(purrr)
library(rlang)
pmap_dfr(df, ~with(list(...), 
               set_names(c(id, cond, 
                           map_dbl(c("A","B","C"),
                                 ~ eval_tidy(parse_expr(paste(.x,cond,sep = "_"))))),
                          c("id","cond","A_def","B_def","C_def"))
               ))
# A tibble: 6 x 5
     id  cond A_def B_def C_def
  <dbl> <dbl> <dbl> <dbl> <dbl>
1     1     1     6     5     6
2     2     1     4     8     1
3     3     2     2     8     2
4     4     2     4     2     5
5     5     3     0     0     8
6     6     3     7     5     4

Answer 3

Here's a short base R solution using mapply :

f <- function(x, i) df[-(1:2)][i, c(x, x+3, x+6)]
df <- cbind(df[1:2], t(mapply(f, df$cond, seq(nrow(df)))))
setNames(df, c("id", "cond", "A_def", "B_def", "C_def"))
#>    id cond A_def B_def C_def
#> 1 001    1     7     8     1
#> 2 002    1     2     5     2
#> 3 003    2     4     2     3
#> 4 004    2     0     3     1
#> 5 005    3     9     0     7
#> 6 006    3     9     7     0