fct_lump_n warning: one observation per factor level

[MagnusNordmo]

fct_lump_n() relies on frequency information to identify the top n levels, but with only one observation per factor level, it may not produce meaningful results. This can occur in various contexts and often reflects a misspecified request. A warning could help users recognize when their data structure is unsuitable for fct_lump_n().

Suggested warning: "warning: only one observation per factor level".

library(tidyverse)
library(reprex)

dat <- tibble(x = sample(c("X","Y","Z","D"),size = 20,replace = T),
       y = sample(1:10,replace = T,size = 20)) 

dat |> 
  mutate(x = fct_lump_n(x,n = 2)) # works
#> # A tibble: 20 × 2
#>    x         y
#>    <fct> <int>
#>  1 Y         8
#>  2 D        10
#>  3 Other     5
#>  4 D         8
#>  5 D         3
#>  6 D         2
#>  7 D         6
#>  8 Y         1
#>  9 Y         5
#> 10 D         8
#> 11 X         6
#> 12 Y         5
#> 13 D         3
#> 14 X         2
#> 15 Y         2
#> 16 D        10
#> 17 D         1
#> 18 X         9
#> 19 X         5
#> 20 X         2

dat |> 
  summarise(mean = mean(y),.by = x) |> 
  mutate(x = fct_lump_n(x,n=2)) # does not work with no warning
#> # A tibble: 4 × 2
#>   x      mean
#>   <fct> <dbl>
#> 1 Y      4.2 
#> 2 D      5.67
#> 3 Z      5   
#> 4 X      4.8

# warning: only one observation per factor level

^{Created on 2024-12-06 with reprex v2.1.0}