Counting Arguments in the Tidyverse

Before we start anything, I’d like to mention that most of the hard work came from nsaunders and his great blog post Idle thoughts lead to R internals: how to count function arguments.

Let’s get started. The aim of this blog is to capture the number of arguments present in each function with packages of the {tidyverse}. First we need to load the necessary packages

library("tidyverse")
library("tidytext")

Now we need to grab the relevant {tidyverse} packages

tpkg = tidyverse_packages()
tpkg[17] = "readxl"
head(tpkg)
## [1] "broom"  "cli"  "crayon"  "dplyr"  "dbplyr"  "forcats"

We’ve had to reset the 17th element to {readxl} as it gets loaded as readxl\n(>=, which breaks the next block of code. Now we also need to load in the tidyverse packages. Doing this one by one would be a pain, so I’ve used map()

map(tpkg, library, character.only = TRUE)

Now for the actual analysis I’m just going to whack the full code in now, then go through it line by line.

pkg = tpkg %>%
  as_tibble() %>%
  rename(package = value) %>%
  rowwise() %>%
  mutate(funcs = paste0(ls(paste0("package:", package)), collapse = ",")) %>%
  unnest_tokens(func, funcs, token = stringr::str_split,
                pattern = ",", to_lower = FALSE) %>%
  filter(is.function(get(func, pos = paste0("package:", package)))) %>%
  mutate(num_args = length(formalArgs(args(get(func, pos = paste0("package:", package)))))) %>%
  ungroup()

This is what the head of pkg looks like

head(pkg)
## # A tibble: 6 x 3
##   package func            num_args
##
## 1 broom   augment                2
## 2 broom   augment_columns        8
## 3 broom   bootstrap              3
## 4 broom   confint_tidy           4
## 5 broom   finish_glance          2
## 6 broom   fix_data_frame         3

Lines `1-4`

Lines 1-4 look like this

tpkg %>%
  as_tibble() %>%
  rename(package = value) %>%
  rowwise() %>%

Here we are grabbing, the tidyverse packages character vector, converting it to a tibble and renaming the column. We then use rowwise() so that we can work in a row-wise fashion.

Line 5

mutate(funcs = paste0(ls(paste0("package:", package)), collapse = ",")) %>%

To get a character vector back of the objects within a package, we do ls("package:package_name"). However, we want to store this as a single string, so we need to use our old friend paste0() to do so. We then use mutate to attach this to the data frame. Our data from now looks like this

## Source: local data frame [6 x 2]
## Groups:
##
## # A tibble: 6 x 2
##   package funcs
##
## 1 broom   argument_glossary,augment,augment_columns,bootstrap,column_gloss…
## 2 cli     ansi_hide_cursor,ansi_show_cursor,ansi_with_hidden_cursor,bg_bla…
## 3 crayon  %+%,bgBlack,bgBlue,bgCyan,bgGreen,bgMagenta,bgRed,bgWhite,bgYell…
## 4 dplyr   %>%,add_count,add_count_,add_row,add_rownames,add_tally,add_tall…
## 5 dbplyr  add_op_single,as.sql,base_agg,base_no_win,base_odbc_agg,base_odb…
## 6 forcats %>%,as_factor,fct_anon,fct_c,fct_collapse,fct_count,fct_cross,fc…

Lines 6 - 7

unnest_tokens(func, funcs, token = stringr::str_split,
                pattern = ",", to_lower = FALSE) %>%

As we’ve stored the function names as a single string, we can now apply some {tidytext} to turn our data into long data! We do this using the unnest_tokens() function. Here we are taking the funcs variable, turning it into func by splitting it up using str_split() from {stringr}. The data now looks like this

## Source: local data frame [6 x 2]
## Groups:
##
## # A tibble: 6 x 2
##   package func
##
## 1 broom   argument_glossary
## 2 broom   augment
## 3 broom   augment_columns
## 4 broom   bootstrap
## 5 broom   column_glossary
## 6 broom   confint_tidy

Line 8

filter(is.function(get(func, pos = paste0("package:", package)))) %>%

Now, not every object inside a package is a function. We can use is.function() to test this. However, as our function names are stored as strings, we must wrap them in the get() function. For instance,

is.function("augment")
## [1] FALSE
is.function(get("augment"))
## [1] TRUE

What if we have conflicts in function names? We can also specify the package our function is from, using the argument pos

is.function(get("augment", pos = "package:broom"))
## [1] TRUE

We can then use this condition within a filter command to remove any objects that aren’t functions

Lines 9 - end

mutate(num_args = length(formalArgs(args(get(func, pos = paste0("package:", package)))))) %>%
  ungroup()

It is possible to withdraw the arguments of a function using the formalArgs() function. However, this does not work on primitive functions

formalArgs(get("add", pos = "package:magrittr"))
## NULL
formalArgs(get("augment", pos = "package:broom"))
## [1] "x"   "..."

We can counter act this by wrapping the function in args() first. This method now works for both primitives and non-primitives

formalArgs(args(get("add", pos = "package:magrittr")))
## [1] "e1" "e2"
formalArgs(args(get("augment", pos = "package:broom")))
## [1] "x"   "..."

To work out the number of these argument we simply wrap this expression in length().

The big reveal

pkg %>%
  arrange(desc(num_args))
## # A tibble: 2,292 x 3
##    package    func               num_args
##
##  1 ggplot2    theme                    93
##  2 ggplot2    guide_colorbar           28
##  3 ggplot2    guide_colourbar          28
##  4 ggplot2    guide_legend             21
##  5 rstudioapi launcherSubmitJob        21
##  6 ggplot2    geom_dotplot             19
##  7 ggplot2    geom_boxplot             18
##  8 readr      read_delim_chunked       18
##  9 readr      read_delim               17
## 10 readr      spec_delim               17
## # … with 2,282 more rows

So it turns out that theme() from {ggplot2} is king of the arguments, by a mile! The largest per package looks like this

We’re not done there! The 9 packages with the largest sum of arguments are

largest = pkg %>%
  group_by(package) %>%
  count() %>%
  arrange(desc(n)) %>%
  head(9) %>%
  pull(package)
largest
## [1] "rlang"      "ggplot2"    "dplyr"      "purrr"      "lubridate"
## [6] "dbplyr"     "readr"      "rstudioapi" "httr"

We can plot a histogram, for each package, of the no. of arguments in each function like so..

pkg %>%
  filter(package %in% largest) %>%
  ggplot(aes(x = num_args)) +
  geom_histogram(binwidth = 1, fill = "steelblue", col = "black") +
  facet_wrap(~package) +
  xlim(c(0, 25)) +
  theme_minimal()

We can go a step further and retrieve the argument names as well. To do this we use the same technique as before with the functions

pkg %>%
  rowwise() %>%
  mutate(args = paste0(formalArgs(args(get(func, pos = paste0("package:", package)))),
                       collapse = ",")) %>%
  unnest_tokens(arg, args, token = stringr::str_split,
                pattern = ",", to_lower = FALSE) %>%
  ungroup() %>%
  count(arg) %>%
  arrange(desc(n))
## # A tibble: 1,029 x 2
##    arg          n
##
##  1 ...        785
##  2 x          698
##  3 data       169
##  4 .x         120
##  5 ""         102
##  6 n           91
##  7 .f          90
##  8 position    90
##  9 mapping     79
## 10 na.rm       79
## # … with 1,019 more rows

The most commonly used arguments in the tidyverse are ... and x by some distance.

pkg %>%
  rowwise() %>%
  mutate(args = paste0(formalArgs(args(get(func, pos = paste0("package:", package)))),
                       collapse = ",")) %>%
  unnest_tokens(arg, args, token = stringr::str_split,
                pattern = ",", to_lower = FALSE) %>%
  ungroup() %>%
  group_by(package) %>%
  count(arg) %>%
  arrange(package, desc(n)) %>%
  slice(2) %>%
  arrange(desc(n))
## # A tibble: 26 x 3
## # Groups:   package [26]
##    package   arg         n
##
##  1 ggplot2   data      103
##  2 purrr     .x         91
##  3 dplyr     x          83
##  4 rlang     ...        64
##  5 readr     locale     44
##  6 lubridate ...        35
##  7 stringr   pattern    23
##  8 dbplyr    x          21
##  9 httr      ...        21
## 10 tidyr     ...        18
## # … with 16 more rows

So you can see that data is the most common argument within {ggplot2}, .x is the most common argument within {purrr} and so on…

That’s it for this blog post. Hope you’ve enjoyed!