group_by dplyr within a function NSE

Question

I am having problems using dplyr and group_by within a piped function call.

Reproducible Example:

Using the following data:

ex_data<- structure(list(word1 = c("no", "not", "not", "no", "not", "not", 
"not", "not", "no", "not", "no", "not", "not", "not", "no", "not", 
"no", "no", "not", "not", "not", "no", "not", "without", "never", 
"no", "not", "no", "no", "not", "not", "not", "no", "no", "no", 
"not", "not", "without", "never", "no", "not", "not", "not", 
"not", "not", "never", "no", "no", "not", "not"), word2 = c("doubt", 
"like", "help", "no", "want", "wish", "allow", "care", "harm", 
"sorry", "great", "leave", "pretend", "worth", "pleasure", "love", 
"danger", "want", "afraid", "doubt", "fail", "good", "forget", 
"feeling", "forget", "matter", "avoid", "chance", "hope", "forgotten", 
"miss", "perfectly", "bad", "better", "opportunity", "admit", 
"fair", "delay", "failed", "wish", "dislike", "distress", "refuse", 
"regret", "trust", "want", "evil", "greater", "better", "blame"
), score = c(-1L, 2L, 2L, -1L, 1L, 1L, 1L, 2L, -2L, -1L, 3L, 
-1L, -1L, 2L, 3L, 3L, -2L, 1L, -2L, -1L, -2L, 3L, -1L, 1L, -1L, 
1L, -1L, 2L, 2L, -1L, -2L, 3L, -3L, 2L, 2L, -1L, 2L, -1L, -2L, 
1L, -2L, -2L, -2L, -2L, 1L, 1L, -3L, 3L, 2L, -2L), n = c(102L, 
99L, 82L, 60L, 45L, 39L, 36L, 23L, 22L, 21L, 19L, 18L, 18L, 17L, 
16L, 16L, 15L, 15L, 15L, 14L, 14L, 13L, 13L, 13L, 12L, 12L, 12L, 
11L, 11L, 10L, 10L, 10L, 9L, 9L, 9L, 9L, 9L, 9L, 8L, 8L, 8L, 
8L, 8L, 8L, 8L, 7L, 7L, 7L, 7L, 7L), contribution = c(-102L, 
198L, 164L, -60L, 45L, 39L, 36L, 46L, -44L, -21L, 57L, -18L, 
-18L, 34L, 48L, 48L, -30L, 15L, -30L, -14L, -28L, 39L, -13L, 
13L, -12L, 12L, -12L, 22L, 22L, -10L, -20L, 30L, -27L, 18L, 18L, 
-9L, 18L, -9L, -16L, 8L, -16L, -16L, -16L, -16L, 8L, 7L, -21L, 
21L, 14L, -14L)), .Names = c("word1", "word2", "score", "n", 
"contribution"), row.names = c(NA, -50L), class = c("tbl_df", 
"tbl", "data.frame"))

The regular typical piping operation works as expected:

outside_result<- ex_data %>% 
  mutate(word2=reorder(word2,contribution)) %>% 
  group_by(word1) %>% 
  top_n(10,abs(contribution)) %>% 
  group_by(word1,word2) %>% 
  arrange(desc(contribution)) %>% 
  ungroup() %>% 
  mutate(word2 = factor(paste(word2,word1, sep = "__"),
                              levels=rev(paste(word2,word1,sep="__"))))

I have implemented the above into a function as given below:

order_bars <- function(df,facetPanel,barCategory,value){
        df %>% mutate(barCategory=reorder(barCategory,value)) %>% 
          group_by(facetPanel) %>% 
          top_n(10,abs(value)) %>% 
          group_by(facetPanel,barCategory) %>% 
          arrange(desc(value)) %>% 
          ungroup() %>% 
          mutate(barCategory = factor(paste(barCategory,facetPanel, sep = "__"),
                                     levels=rev(paste(barCategory,facetPanel,sep="__"))))
      }

And taking the advice from this post, used $ notation when referencing variables of a data.frame during mutate operations within a function.

inside_result<-order_bars(ex_data,ex_data$word1,ex_data$word2,ex_data$contribution)

R throws the following error:

Error: unknown variable to group by : facetPanel
Called from: resolve_vars(new_groups, tbl_vars(.data))

I suspect group_by needs to be tweaked to take either named variables, or I have to use .dot notation to reference columns, although I'm just throwing this out into the wind...

Axeman · Accepted Answer · 2016-12-28T13:38:39.360

You'll need to learn how to use 1) the SE versions of dplyr verbs such as group_by_ and mutate_ and 2) the enigmatic lazyeval::interp. Please read vignette("nse") carefully.

Then we can come to:

order_bars <- function(df, facetPanel, barCategory, value){
  require(lazyeval)
  df %>% 
    mutate_(barCategory = interp(~reorder(x, y), x = as.name(barCategory), 
                                 y = as.name(value))) %>% 
    group_by_(facetPanel) %>% 
    filter_(interp(~min_rank(desc(abs(x))) <= 10, x = as.name(value))) %>% 
    group_by_(facetPanel, barCategory) %>% 
    arrange_(interp(~desc(x), x = as.name(value))) %>% 
    ungroup() %>% 
    mutate_(barCategory = interp(
      ~factor(paste(x, y, sep = "__"), levels = rev(paste(x, y, sep = "__"))),
      x = as.name(barCategory), y = as.name(facetPanel)))
}

order_bars(ex_data, 'word1', 'word2', 'contribution')

# A tibble: 25 × 6
   word1    word2 score     n contribution  barCategory
   <chr>    <chr> <int> <int>        <int>       <fctr>
1    not     like     2    99          198    like__not
2    not     help     2    82          164    help__not
3     no    great     3    19           57    great__no
4     no pleasure     3    16           48 pleasure__no
5    not     love     3    16           48    love__not
6    not     care     2    23           46    care__not
7    not     want     1    45           45    want__not
8    not     wish     1    39           39    wish__not
9     no     good     3    13           39     good__no
10   not    allow     1    36           36   allow__not

Note that we need to replace top_n with a filter_ statement, since no top_n_ exists. Looking at the source of top_n makes it obvious how the filter_ statement should be constructed.

Or if you want to get fancy, you can write a NSE version of order_bars:

order_bars <- function(df,facetPanel,barCategory,value){
  facetPanel <- substitute(facetPanel)
  barCategory <- substitute(barCategory)
  value <- substitute(value)

  require(lazyeval)
  df %>% 
    mutate_(barCategory = interp(~reorder(x, y), x = barCategory, y = value)) %>% 
    group_by_(facetPanel) %>% 
    filter_(interp(~min_rank(desc(abs(x))) <= 10, x = value)) %>% 
    group_by_(facetPanel, barCategory) %>% 
    arrange_(interp(~desc(x), x = value)) %>% 
    ungroup() %>% 
    mutate_(barCategory = interp(
      ~factor(paste(x, y, sep = "__"), levels = rev(paste(x, y, sep = "__"))),
      x = barCategory, y = facetPanel))
}

order_bars(ex_data, word1, word2, contribution)

Ideally, you would write only the SE version fully, and link the NSE version to the SE version with lazyeval. I'll leave that as an exercise to the reader.

this is the first time I've run up against SE and it's a good learning opportunity... thanks for the direction! — danny_C_O_T_W, Dec 28 '16 at 13:38
awwww nice answer axeman but man, that SE evaluation stuff is horrible — ℕʘʘḆḽḘ, Dec 28 '16 at 13:53
@Noobie, Yeah, one pays for the prettiness (in my opinion) of NSE with the ugliness (in my opinion) of SE when using the tidyverse. — Axeman, Dec 28 '16 at 13:56
This behaviour has been changed. "dplyr now uses tidy evaluation semantics. NSE verbs still capture their arguments, but you can now unquote parts of these arguments. This offers full programmability with NSE verbs. Thus, the underscored versions are now superfluous." See [programming with dplyr](https://cran.r-project.org/web/packages/dplyr/vignettes/programming.html) — Paul Rougieux, Oct 11 '18 at 08:36

score 1 · Answer 2 · answered Jul 07 '19 at 15:52

With rlang_0.4.0 and dplyr_0.8.2, we can use the tidy-evaluation operator ({{...}}) or curly-curly which abstracts quote-and-unquote into a single interpolation step.

library(rlang)
library(dplyr)
order_barsN <- function(df, facetPanel, barCategory, value) {
    df %>% 
        mutate(barCategory = reorder({{barCategory}}, {{value}}))%>%
        group_by({{facetPanel}}) %>%
        filter(min_rank(desc(abs({{value}}))) <= 10) %>%
        group_by({{facetPanel}}, {{barCategory}}) %>%
        arrange(desc({{value}})) %>%
        ungroup %>%
        mutate(barCategory = factor(str_c({{barCategory}}, {{facetPanel}}, sep="__"),
                levels = rev(str_c({{barCategory}}, {{facetPanel}}, sep="__"))))

        }


out2 <- order_barsN(ex_data, word1, word2, contribution)

-checking with previous answer

out1 <- order_bars(ex_data, word1, word2, contribution)
identical(out1, out2)
#[1] TRUE

group_by dplyr within a function NSE

2 Answers2

Linked