CODEWITHSUNDEEP
rdplyrrlangquasiquotes
Indrajeet Patil
Indrajeet Patil

Reputation: 4879

using `rlang` quasiquotation with `dplyr::_join` functions

I am trying to write a custom function where I use rlang's quasiquotation. This function also internally uses dplyr's join functions. I have provided below a minimal working example that illustrated my problem. 

# needed libraries 
library(tidyverse)

# function definition
df_combiner <- function(data, x, group.by) {
  # check how many variables were entered for this grouping variable
  group.by <- as.list(rlang::quo_squash(rlang::enquo(group.by)))

  # based on number of arguments, select `group.by` in cases like `c(cyl)`,
  # the first list element after `quo_squash` will be `c` which we don't need,
  # but if we pass just `cyl`, there is no `c`, this will take care of that
  # issue
  group.by <-
    if (length(group.by) == 1) {
      group.by
    } else {
      group.by[-1]
    }

  # creating internal dataframe
  df <- dplyr::group_by(.data = data, !!!group.by, .drop = TRUE)

  # creating dataframes to be joined: one with tally, one with summary
  df_tally <- dplyr::tally(df)
  df_mean <- dplyr::summarise(df, mean = mean({{ x }}, na.rm = TRUE))

  # without specifying `by` argument, this works but prints a message I want to avoid
  print(dplyr::left_join(x = df_tally, y = df_mean))

  # joining by specifying `by` argument (my failed attempt)
  dplyr::left_join(x = df_tally, y = df_mean, by = !!!group.by)
}

# using the function
df_combiner(diamonds, carat, c(cut, clarity))

#> Joining, by = c("cut", "clarity")

#> # A tibble: 40 x 4
#> # Groups:   cut [5]
#>    cut   clarity     n  mean
#>    <ord> <ord>   <int> <dbl>
#>  1 Fair  I1        210 1.36 
#>  2 Fair  SI2       466 1.20 
#>  3 Fair  SI1       408 0.965
#>  4 Fair  VS2       261 0.885
#>  5 Fair  VS1       170 0.880
#>  6 Fair  VVS2       69 0.692
#>  7 Fair  VVS1       17 0.665
#>  8 Fair  IF          9 0.474
#>  9 Good  I1         96 1.20 
#> 10 Good  SI2      1081 1.04 
#> # ... with 30 more rows

#> Error in !group.by: invalid argument type

As can be seen here, I want to avoid the message #> Joining, by = c("cut", "clarity") and so explicitly want to input the by argument for the _join function but I am not sure how to do this. (I've tried rlang::as_string, rlang::quo_name, etc.).

Upvotes: 4

Views: 441

Answers (4)

MrFlick
MrFlick

Reputation: 206167

With the latest versions of dplyr (>=1.1.0), you use join_by to build your merge. Here we swap out the join line with

dplyr::left_join(x = df_tally, y = df_mean, join_by(!!!group.by))

Since the value of group.by is a list, the !!! operator will expand that into arguments so join_by(!!!group.by) will be expanded to join_by(cut, clarity).

More generally, say you are doing a more simple join with

a <- data.frame(ID=1:3, letter=letters[1:3])
b <- data.frame(id_col=3:1, state=state.abb[1:3])
dplyr::inner_join(a, b, join_by(ID==id_col))

If you want to move that to a function you could pass in the whole join_by clause

foo <- function(x, y, join) {
  dplyr::inner_join(x, y, join)  
}
foo(a, b, join_by( ID == id_col ))

or you can pass in individual column names

foo <- function(x, y, x_col, y_col) {
  dplyr::inner_join(x, y, join_by( {{x_col}} == {{y_col}} ))
}
foo(a, b, ID, id_col)

Upvotes: 1

Silviculturalist
Silviculturalist

Reputation: 199

As mentioned by earlier authors, ´by´ expects a string vector. An easy way to move from lists of quosures to strings is illustrated by stanwood on the RStudio Community thread Should tidyeval be abandoned?

...tidyr::left_join still expects a list of strings: by = c("Species", "Sepal.Length"). If I want to supply these programatically the best solution I found was by = sapply(sepaldims, quo_text). Consider this a plug for abstracting quo_text to lists of quosures.

sepaldims <- quos(Species, Sepal.Length)

Upvotes: 1

akrun
akrun

Reputation: 886928

We can convert to string with as_string

dplyr::left_join(x = df_tally, y = df_mean,
            by = map_chr(group.by, rlang::as_string))

df_combiner <- function(data, x, group.by) {
  # check how many variables were entered for this grouping variable
  group.by <- as.list(rlang::quo_squash(rlang::enquo(group.by)))

  # based on number of arguments, select `group.by` in cases like `c(cyl)`,
  # the first list element after `quo_squash` will be `c` which we don't need,
  # but if we pass just `cyl`, there is no `c`, this will take care of that
  # issue
  group.by <-
    if (length(group.by) == 1) {
      group.by
    } else {
      group.by[-1]
    }

  # creating internal dataframe
  df <- dplyr::group_by(.data = data, !!!group.by, .drop = TRUE)

  # creating dataframes to be joined: one with tally, one with summary
  df_tally <- dplyr::tally(df)
  df_mean <- dplyr::summarise(df, mean = mean({{ x }}, na.rm = TRUE))

  # without specifying `by` argument, this works but prints a message I want to avoid
  #print(dplyr::left_join(x = df_tally, y = df_mean))

  # joining by specifying `by` argument (my failed attempt)
   dplyr::left_join(x = df_tally, y = df_mean, by = map_chr(group.by, rlang::as_string))

}

-checking

df_combiner(diamonds, carat, c(cut, clarity))
# A tibble: 40 x 4
# Groups:   cut [5]
#   cut   clarity     n  mean
#   <ord> <ord>   <int> <dbl>
# 1 Fair  I1        210 1.36 
# 2 Fair  SI2       466 1.20 
# 3 Fair  SI1       408 0.965
# 4 Fair  VS2       261 0.885
# 5 Fair  VS1       170 0.880
# 6 Fair  VVS2       69 0.692
# 7 Fair  VVS1       17 0.665
# 8 Fair  IF          9 0.474
# 9 Good  I1         96 1.20 
#10 Good  SI2      1081 1.04 
# … with 30 more rows

Upvotes: 5

Artem Sokolov
Artem Sokolov

Reputation: 13691

Join functions take a string vector for its by argument. Use deparse to go from expressions to strings:

dplyr::left_join(x = df_tally, y = df_mean, by = map_chr(group.by, deparse))

Upvotes: 3

Related Questions