boxplot.stats in dplyr with groups

user_882102 0
Viewed 2045

I need to calculate boxplot statistics for a data.frame with many, many groups.

What I ideally need is:

library(dplyr)
iris %>%
  group_by(Species) %>%
  summarise(boxplot=boxplot.stats(Sepal.Length))) # + some kind of magic

# A tibble: 3 x 6
  Species    lower_whisker lower_hinge median upper_hinge upper_whisker
  <fct>              <dbl>       <dbl>  <dbl>       <dbl>         <dbl>
1 setosa               4.3         4.8    5           5.2           5.8
2 versicolor           4.9         5.6    5.9         6.3           7  
3 virginica            5.6         6.2    6.5         6.9           7.9

But so far, I've managed to do a half-purrr mapping thing and cannot unpack it.

boxplot.stats2 <- function(x, ...) {
  res <- boxplot.stats(x, ...)
  res <- res$stats
  names(res) <- c('lower_whisker','lower_hinge','median','upper_hinge','upper_whisker')
  #t(as.data.frame(res))
  res
}

iris %>%
  group_by(Species) %>%
  summarise(boxplot=list(boxplot.stats2(Sepal.Length)),
    #manual unpacking
    lower_whisker = boxplot[[1]]['lower_whisker'],
    lower_hinge = boxplot[[1]]['lower_hinge'],
    median = boxplot[[1]]['median'],
    upper_hinge = boxplot[[1]]['upper_hinge'],
    upper_whisker = boxplot[[1]]['upper_whisker']
  )

It gives the same result, but I suspect there should be a more elegant solution for it.

purrr dplyr r
1 Answers

If you use a list as your summary output you can use the unnest() functions from package tidyr.

Newer versions of tidyr have some new functions, including unnest_wider(), which is a handy tool for your situation.

Here I'll get only the 5 numbers of the boxplot stats and put them in a list in summarise(), much as you started to do. I also name them, since the stats from boxplot.stats() don't have any identifying info.

The new column is a list column, containing a named vector of 5 values for each species.

library(dplyr)
library(tidyr) # development version, tidyr_0.8.3.9000

iris %>%
    group_by(Species) %>%
    summarise(boxplot= list( setNames(boxplot.stats(Sepal.Length)$stats,
                                      c('lower_whisker','lower_hinge','median','upper_hinge','upper_whisker') ) ) )

# A tibble: 3 x 2
  Species    boxplot  
  <fct>      <list>   
1 setosa     <dbl [5]>
2 versicolor <dbl [5]>
3 virginica  <dbl [5]>

Following that up with unnest_wider() gives you the desired results.

iris %>%
    group_by(Species) %>%
    summarise(boxplot= list( setNames(boxplot.stats(Sepal.Length)$stats,
                                      c('lower_whisker','lower_hinge','median','upper_hinge','upper_whisker') ) ) ) %>%
    unnest_wider(boxplot)

# A tibble: 3 x 6
  Species    lower_whisker lower_hinge median upper_hinge upper_whisker
  <fct>              <dbl>       <dbl>  <dbl>       <dbl>         <dbl>
1 setosa               4.3         4.8    5           5.2           5.8
2 versicolor           4.9         5.6    5.9         6.3           7  
3 virginica            5.6         6.2    6.5         6.9           7.9

You can skip the naming step, but then you'd need to name the columns at the end.

iris %>%
    group_by(Species) %>%
    summarise(boxplot= list( boxplot.stats(Sepal.Length)$stats ) ) %>%
    unnest_wider(boxplot)

# A tibble: 3 x 6
  Species     ...1  ...2  ...3  ...4  ...5
  <fct>      <dbl> <dbl> <dbl> <dbl> <dbl>
1 setosa       4.3   4.8   5     5.2   5.8
2 versicolor   4.9   5.6   5.9   6.3   7  
3 virginica    5.6   6.2   6.5   6.9   7.9

Older versions of tidyr (pre 1.0.0) could still help here, although the work would take quite a few more steps. Since list names are lost in the current unnest() you would need to add these manually before unnesting so you can spread() into new columns.

iris %>%
group_by(Species) %>%
summarise(boxplot= list( boxplot.stats(Sepal.Length)$stats),
          stat = list( c('lower_whisker','lower_hinge','median','upper_hinge','upper_whisker') ) ) %>%
unnest(stat, boxplot) %>%
spread(stat, boxplot)

# A tibble: 3 x 6
  Species    lower_hinge lower_whisker median upper_hinge upper_whisker
  <fct>            <dbl>         <dbl>  <dbl>       <dbl>         <dbl>
1 setosa             4.8           4.3    5           5.2           5.8
2 versicolor         5.6           4.9    5.9         6.3           7  
3 virginica          6.2           5.6    6.5         6.9           7.9
user_24615520
Related