R data.table 按组排序，每组底部有“other”

Question

I can't quite get the syntax right for this. I have a data.table where I would like to sort first by a grouping column g1 (ordered factor), then in descending order by another column n . The only catch is that I would like rows labeled "other" for a third column g2 to appear at the bottom of each group, regardless of their value of n .

Example:

library(data.table)

dt <- data.table(g1 = factor(rep(c('Australia', 'Mexico', 'Canada'), 3), levels = c('Australia', 'Canada', 'Mexico')),
                 g2 = rep(c('stuff', 'things', 'other'), each = 3),
                 n = c(1000, 2000, 3000, 5000, 100, 3500, 10000, 10000, 0))

This is the expected output, where within each g1 , we have descending order of n except that rows where g2 == 'other' are always at the bottom:

         g1     g2     n
1: Australia things  5000
2: Australia  stuff  1000
3: Australia  other 10000
4:    Canada things  3500
5:    Canada  stuff  3000
6:    Canada  other     0
7:    Mexico  stuff  2000
8:    Mexico things   100
9:    Mexico  other 10000

Answer 1

Take advantage of data.table::order and its - -reverse ordering:

dt[order(g1, g2 == "other", -n), ]
#           g1     g2     n
#       <fctr> <char> <num>
# 1: Australia things  5000
# 2: Australia  stuff  1000
# 3: Australia  other 10000
# 4:    Canada things  3500
# 5:    Canada  stuff  3000
# 6:    Canada  other     0
# 7:    Mexico  stuff  2000
# 8:    Mexico things   100
# 9:    Mexico  other 10000

We add g2 == "other" because you said that "other" should always be last. If, for example, "stuff" was "abc" , then we can see the difference in behavior:

dt[ g2 == "stuff", g2 := "abc" ]
dt[order(g1, -n), ]
#           g1     g2     n
#       <fctr> <char> <num>
# 1: Australia  other 10000
# 2: Australia things  5000
# 3: Australia    abc  1000
# 4:    Canada things  3500
# 5:    Canada    abc  3000
# 6:    Canada  other     0
# 7:    Mexico  other 10000
# 8:    Mexico    abc  2000
# 9:    Mexico things   100

dt[order(g1, g2 == "other", -g2), ]
#           g1     g2     n
#       <fctr> <char> <num>
# 1: Australia things  5000
# 2: Australia    abc  1000
# 3: Australia  other 10000
# 4:    Canada things  3500
# 5:    Canada    abc  3000
# 6:    Canada  other     0
# 7:    Mexico things   100
# 8:    Mexico    abc  2000
# 9:    Mexico  other 10000

One disadvantage of this is that setorder doesn't work directly:

setorder(dt, g1, g2 == "other", -n)
# Error in setorderv(x, cols, order, na.last) : 
#   some columns are not in the data.table: ==,other

so we instead need to reorder and reassign back to dt .

BTW: this works because g2 == "other" resolves to logical , yes, but in sorting those are treated as 0 (false) and 1 (true), ergo false-conditions will appear before true-conditions.