How to sort a dataframe in R by one variable while grouping for others

Using base R, sorted by the 'var_3' values doubled by 'var_2'.

with(test_frame, test_frame[order(-rep(var_3[!duplicated(var_2)], each=2)), ])
# # A tibble: 10 x 3
#    var_1 var_2 var_3
#    <chr> <chr> <dbl>
#  1 a     a4     4.79
#  2 b     a4     1.33
#  3 a     a2     3.24
#  4 b     a2     4.62
#  5 a     a5     3.06
#  6 b     a5     2.56
#  7 a     a3     1.55
#  8 b     a3     4.96
#  9 a     a1     1.47
# 10 b     a1     2.90

Data

test_frame <- structure(list(var_1 = c("a", "b", "a", "b", "a", "b", "a", "b", 
"a", "b"), var_2 = c("a1", "a1", "a2", "a2", "a3", "a3", "a4", 
"a4", "a5", "a5"), var_3 = c(1.46994944661856, 2.89998832624406, 
3.24133098497987, 4.61612554918975, 1.55484067089856, 4.95556691568345, 
4.78667293023318, 1.32975023239851, 3.05684713739902, 2.56081386841834
)), row.names = c(NA, -10L), class = c("tbl_df", "tbl", "data.frame"
))

One dplyr option could be:

test_frame %>%
 mutate(ranking = dense_rank(desc((var_1 == "a") * var_3))) %>%
 group_by(var_2) %>%
 mutate(ranking = min(ranking)) %>%
 arrange(ranking) %>%
 select(-ranking)

   var_1 var_2 var_3
   <chr> <chr> <dbl>
 1 a     a4     4.46
 2 b     a4     2.68
 3 a     a5     2.80
 4 b     a5     2.65
 5 a     a1     1.91
 6 b     a1     2.99
 7 a     a3     1.22
 8 b     a3     1.93
 9 a     a2     1.10
10 b     a2     4.92

Or:

test_frame %>%
 filter(var_1 == "a") %>%
 mutate(ranking = dense_rank(desc(var_3))) %>%
 bind_rows(test_frame %>%
            filter(var_1 == "b")) %>%
 group_by(var_2) %>%
 mutate(ranking = min(ranking, na.rm = TRUE)) %>%
 arrange(ranking) %>%
 select(-ranking)

One solution is to pivot_wider so you are sorting on a complete variable, then sort, then pivot_longer back to the original shape.

test_frame %>%
 pivot_wider( names_from = var_1, values_from = var_3) %>%
 arrange( -a) %>%
 pivot_longer(cols=c(a,b), names_to="var_1", values_to = "var_3")

# A tibble: 10 x 3
   var_2 var_1 var_3
   <chr> <chr> <dbl>
 1 a1    a      4.21
 2 a1    b      1.82
 3 a5    a      3.71
 4 a5    b      1.25
 5 a3    a      2.76
 6 a3    b      2.58
 7 a2    a      2.60
 8 a2    b      4.32
 9 a4    a      1.12
10 a4    b      1.54

No need to group and add and remove columns from your dataframe - it's just good use of the dplyr::arrange() method. I think this gets you what you need:

as.data.frame(test_frame) %>% arrange(var_3, var_1, var_2)

That gives you this:

    var_1 var_2    var_3
1      b    a4 1.866265
2      a    a4 2.703378
3      b    a5 2.931703
4      a    a1 2.935217
5      a    a2 3.019241
6      b    a1 3.029589
7      b    a3 3.657182
8      a    a3 4.392643
9      b    a2 4.415388
10     a    a5 4.498499

The only problem is that var_2 is sorted 'b', then 'a' - not 'a', then 'b' like you wanted. There are probably a few ways around that (you can use desc(...) within the arrange() function... but I had some trouble getting it to work. In the end, you can actually separate out the arrange() functions, which kind of acts to sort each column in a specific order. Here's my final solution for you:

as.data.frame(test_frame) %>% arrange(var_3) %>% arrange(var_1) %>% arrange(var_2)

   var_1 var_2    var_3
1      a    a1 2.935217
2      b    a1 3.029589
3      a    a2 3.019241
4      b    a2 4.415388
5      a    a3 4.392643
6      b    a3 3.657182
7      a    a4 2.703378
8      b    a4 1.866265
9      a    a5 4.498499
10     b    a5 2.931703