CODEWITHSUNDEEP
rdplyr
Blaze Tama
Blaze Tama

Reputation: 10948

Get top values of multiple group bys

I've been trying a few ways to achieve (do, row_number) this but still stuck.

I have 3 groups: month, city, and gender.

I would like to get only the top 5 count of these 3 group bys.

This code works fine only with 2 groups:

df_top5_2grp <- df %>%
            group_by(month, city) %>%
            tally() %>%
            top_n(n = 5, wt = n) %>%
            arrange(retention_month, desc(n))

However, it won't return the top 5 count if I add an additional group:

df_top5_3grp <- df %>%
            group_by(month, city, gender) %>%
            tally() %>%
            top_n(n = 5, wt = n) %>%
            arrange(retention_month, gender, desc(n))

It returns all rows instead. The only difference is I added gender.

Any help is appreciated. Thanks!

Upvotes: 0

Views: 74

Answers (1)

Jon Spring
Jon Spring

Reputation: 66490

You probably need an ungroup() in there.

In the first example below, it returns all the rows, since there are 7 groups, each with one row. So returning the top 5 of each of the seven groups returns all rows.

 mtcars %>%
   group_by(cyl, vs, am) %>%    # grouping across three variables
   tally() %>%                  # tally is a summarization that removes the last grouping
   top_n(n = 5, wt = n)
# A tibble: 7 x 4
# Groups:   cyl, vs [5]      # NOTE! This reminds us the data is still grouped
    cyl    vs    am     n
  <dbl> <dbl> <dbl> <int>
1     4     0     1     1
2     4     1     0     3
3     4     1     1     7
4     6     0     1     3
5     6     1     0     4
6     8     0     0    12
7     8     0     1     2

Adding ungroup makes it so the top 5 filtering happens across all the summarized groups, not within each group.

 mtcars %>%
   group_by(cyl, vs, am) %>%
   tally() %>% 
   ungroup() %>%
   top_n(n = 5, wt = n)
# A tibble: 5 x 4
    cyl    vs    am     n
  <dbl> <dbl> <dbl> <int>
1     4     1     0     3
2     4     1     1     7
3     6     0     1     3
4     6     1     0     4
5     8     0     0    12

Upvotes: 1

Related Questions