dplyr - Group by and select TOP x %

12
votes

Using the package dplyr and the function sample_frac it is possible to sample a percentage from every group. What I need is to first sort the elements in every group and then select top x% from every group?

There is a function top_n, but here I can only determine the number of rows, and I need a relative value.

For example the following data is grouped by gear and sorted by wt within each group:

library(dplyr)
mtcars %>%
  select(gear, wt) %>%
  group_by(gear) %>%
  arrange(gear, wt)

    gear    wt
1   3   2.465
2   3   3.215
3   3   3.435
4   3   3.440
5   3   3.460
6   3   3.520
7   3   3.570
8   3   3.730
9   3   3.780
10  3   3.840
11  3   3.845
12  3   4.070
13  3   5.250
14  3   5.345
15  3   5.424
16  4   1.615
17  4   1.835
18  4   1.935
19  4   2.200
20  4   2.320
21  4   2.620
22  4   2.780
23  4   2.875
24  4   3.150
25  4   3.190
26  4   3.440
27  4   3.440
28  5   1.513
29  5   2.140
30  5   2.770
31  5   3.170
32  5   3.570

Now I would like to select top 20 % within each gear group.

It would be very nice if the solution could be integrated with dplyr's group_by function.

5
rdplyr
Please provide a minimal reproducible example - Steven Beaupré
Couldn't you just calculate the percentage yourself? I'm not sure if this works since we don't have a reproducible example, but I think it might: my_data %>% group_by(my_var) %>% arrange(my_var) %>% filter(top_n()/n() == x%) - tblznbits
@brittenb Thanks for ypur help! top_n() cannot be used without arguments. - DatamineR

5 Answers

23
votes

Or another option with dplyr:

mtcars %>% select(gear, wt) %>% 
  group_by(gear) %>% 
  arrange(gear, desc(wt)) %>% 
  filter(wt > quantile(wt, .8))

Source: local data frame [7 x 2]
Groups: gear [3]

   gear    wt
  (dbl) (dbl)
1     3 5.424
2     3 5.345
3     3 5.250
4     4 3.440
5     4 3.440
6     4 3.190
7     5 3.570
9
votes

Here's another way

mtcars %>% 
  select(gear, wt) %>% 
  arrange(gear, desc(wt)) %>% 
  group_by(gear) %>% 
  slice(seq(n()*.2))

   gear    wt
  (dbl) (dbl)
1     3 5.424
2     3 5.345
3     3 5.250
4     4 3.440
5     4 3.440
6     5 3.570

I take "top" to mean "having the highest value for wt" and so used desc().

5
votes

I believe this gets to the answer you're looking for.

library(dplyr)

mtcars %>% select(gear, wt) %>% 
  group_by(gear) %>% 
  arrange(gear, wt) %>% 
  filter(row_number() / n() <= .2)
2
votes

I know this is coming late, but might help someone now. dplyr has a new function top_frac

  library(dplyr)
mtcars %>%
  select(gear, wt) %>%
  group_by(gear) %>%
  arrange(gear, wt) %>%
  top_frac(n = 0.2,wt = wt)

Here n is the fraction of rows to return and wt is the variable to be used for ordering.

The output is as below.

gear wt 3 5.250 3 5.345 3 5.424
4 3.440 4 3.440 5 3.570

0
votes

A slight variation using top_n and dplyr:

mtcars %>% 
 group_by(gear) %>% 
 select(gear, wt) %>% 
 arrange(gear) %>% 
 top_n(seq(n()*.2))

  gear    wt
  <dbl> <dbl>
1     3  5.42
2     3  5.34
3     3  5.25
4     4  3.44
5     4  3.44
6     5  3.57