base::ifelse() within dplyr::arrange() for conditional arrangement of grouped rows - r

Question

I'm trying to order the rows of a data.frame conditional upon the value of another column.
Here's an example below:
library(magrittr)
library(dplyr)
df <- data.frame(grp = c(1,1,1,2,2,2),
ori = c("f","f","f","r","r","r"),
ite = c("A","B","C","A","B","C"))
df
# # grp ori ite
# 1 1 f A
# 2 1 f B
# 3 1 f C
# 4 2 r A
# 5 2 r B
# 6 2 r C
df %>%
group_by(grp) %>%
arrange(ifelse(ori == "f", ite, desc(ite)), .by_group = TRUE) %>%
ungroup()
# # A tibble: 6 × 3
# # Groups: grp [2]
# grp ori ite
# <dbl> <chr> <chr>
# 1 1 f A
# 2 1 f B
# 3 1 f C
# 4 2 r A
# 5 2 r B
# 6 2 r C
The expected output is:
# # grp ori ite
# 1 1 f A
# 2 1 f B
# 3 1 f C
# 4 2 r C
# 5 2 r B
# 6 2 r A
I have a general idea of why it doesn't work: arrange() cannot look at things on a per-row basis, which is what the ifelse() is asking it to do.
Is there a better way of accomplishing this?

Answer 1

The idea to use ifelse(ori == "f", ite, desc(ite)) is basically good, unfortunately desc(ite) has a negative numeric vector as output, whereas the output of ite is a character vector.
ifelse(df$ori == "f", df$ite, dplyr::desc(df$ite))
#> [1] "A" "B" "C" "-1" "-3" "-5"
To bring the result of ite in reverse order using the same output as input we can write a function asc() which just does the opposite of desc():
asc <- function(x) {
xtfrm(x)
}
No we can use both inside ifelse():
library(dplyr)
df <- data.frame(grp = c(1,1,1,2,2,2),
ori = c("f","f","f","r","r","r"),
ite = c("A","B","C","A","B","C"))
df %>%
arrange(ori, ifelse(ori == "r", desc(ite), asc(ite)))
#> grp ori ite
#> 1 1 f A
#> 2 1 f B
#> 3 1 f C
#> 4 2 r C
#> 5 2 r B
#> 6 2 r A
Created on 2022-08-21 by the reprex package (v2.0.1)

Answer 2

One possible way is splitting the column ori and creating a function to then combine the results as following:
df %>%
split(.$ori) %>%
map(function(x) {
if ('f' %in% x$ori) {
x %>%
group_by(grp) %>%
arrange(ite, .by_group = TRUE)
}
else {
x %>%
group_by(grp) %>%
arrange(desc(ite), .by_group = TRUE)
}
}) %>%
bind_rows()
# A tibble: 6 x 3
## Groups: grp [2]
# grp ori ite
# <dbl> <chr> <chr>
#1 1 f A
#2 1 f B
#3 1 f C
#4 2 r C
#5 2 r B
#6 2 r A

Answer 3

One option to achieve your desired result would be to make use of split + arrange + bind_rows like so:
library(dplyr)
library(purrr)
df %>%
split(.$ori) %>%
purrr::imap(
~ if (.y == "f") arrange(.x, grp, ite) else arrange(.x, grp, desc(ite))
) %>%
dplyr::bind_rows()
#> grp ori ite
#> 1 1 f A
#> 2 1 f B
#> 3 1 f C
#> 4 2 r C
#> 5 2 r B
#> 6 2 r A
And thanks to the suggestion by #MartinGal we could save the bind_rows step by making use of purrr::imap_dfr:
df %>%
split(.$ori) %>%
purrr::imap_dfr(
~ if (.y == "f") arrange(.x, grp, ite) else arrange(.x, grp, desc(ite))
)

Answer 4

Here's another solution.
df %>%
arrange(
ifelse(ori == 'f', ite, NA),
desc(ifelse(ori != 'f', ite, NA))
)