编写一个在 dplyr::mutate() 内部工作的自定义函数

Question

我正在努力编写一个在dplyr::mutate() 中工作的函数。

由于rowwise() %>% sum() 在大型数据集上相当慢，建议的替代方法是返回到 baseR。我希望简化此过程，如下所示，但在 mutate 函数中传递数据时遇到问题。

require(tidyverse)
#> Loading required package: tidyverse
#I'd like to write a function that works inside mutate and replaces the rowSums(select()).
cars <- as_tibble(cars)

cars %>% 
  mutate(sum = rowSums(select(., speed, dist), na.rm = T))
#> # A tibble: 50 x 3
#>    speed  dist   sum
#>    <dbl> <dbl> <dbl>
#>  1    4.    2.    6.
#>  2    4.   10.   14.
#>  3    7.    4.   11.
#>  4    7.   22.   29.
#>  5    8.   16.   24.
#>  6    9.   10.   19.
#>  7   10.   18.   28.
#>  8   10.   26.   36.
#>  9   10.   34.   44.
#> 10   11.   17.   28.
#> # ... with 40 more rows

#Here is my first attempt.
rowwise_sum <- function(data, ..., na.rm = FALSE) {
  columns <- rlang::enquos(...)

  data %>% 
    select(!!!columns) %>% 
    rowSums(na.rm = na.rm)
}

#Doesnt' work as expected:
cars %>% mutate(sum = rowwise_sum(speed, dist, na.rm = T))
#> Error in mutate_impl(.data, dots): Evaluation error: no applicable method for 'select_' applied to an object of class "c('double', 'numeric')".

#But alone it is creating a vector.
cars %>% rowwise_sum(speed, dist, na.rm = T)
#>  [1]   6  14  11  29  24  19  28  36  44  28  39  26  32  36  40  39  47
#> [18]  47  59  40  50  74  94  35  41  69  48  56  49  57  67  60  74  94
#> [35] 102  55  65  87  52  68  72  76  84  88  77  94 116 117 144 110

#Appears to not be getting the data passed.  Specifying with a dot works.
cars %>% mutate(sum = rowwise_sum(., speed, dist, na.rm = T))
#> # A tibble: 50 x 3
#>    speed  dist   sum
#>    <dbl> <dbl> <dbl>
#>  1    4.    2.    6.
#>  2    4.   10.   14.
#>  3    7.    4.   11.
#>  4    7.   22.   29.
#>  5    8.   16.   24.
#>  6    9.   10.   19.
#>  7   10.   18.   28.
#>  8   10.   26.   36.
#>  9   10.   34.   44.
#> 10   11.   17.   28.
#> # ... with 40 more rows

所以问题变成了如何通过在函数内部传递数据来解决每次都包含一个点的需求？

rowwise_sum2 <- function(data, ..., na.rm = FALSE) {
  columns <- rlang::enquos(...)

  data %>% 
    select(!!!columns) %>% 
    rowSums(., na.rm = na.rm)
}

#Same error
cars %>% mutate(sum = rowwise_sum2(speed, dist, na.rm = T))
#> Error in mutate_impl(.data, dots): Evaluation error: no applicable method for 'select_' applied to an object of class "c('double', 'numeric')".

#Same result
cars %>% rowwise_sum2(speed, dist, na.rm = T)
#>  [1]   6  14  11  29  24  19  28  36  44  28  39  26  32  36  40  39  47
#> [18]  47  59  40  50  74  94  35  41  69  48  56  49  57  67  60  74  94
#> [35] 102  55  65  87  52  68  72  76  84  88  77  94 116 117 144 110

#Same result
cars %>% mutate(sum = rowwise_sum2(., speed, dist, na.rm = T))
#> # A tibble: 50 x 3
#>    speed  dist   sum
#>    <dbl> <dbl> <dbl>
#>  1    4.    2.    6.
#>  2    4.   10.   14.
#>  3    7.    4.   11.
#>  4    7.   22.   29.
#>  5    8.   16.   24.
#>  6    9.   10.   19.
#>  7   10.   18.   28.
#>  8   10.   26.   36.
#>  9   10.   34.   44.
#> 10   11.   17.   28.
#> # ... with 40 more rows

由reprex package (v0.2.0) 于 2018 年 5 月 22 日创建。

---

下面 akrun 的回答（请点赞）：

套用一句话：放弃 mutate() 并在新函数中执行所有操作。

这是我作为对他的更新的最后一个函数，如果需要，它还允许命名总和值列。

rowwise_sum <- function(data, ..., sum_col = "sum", na.rm = FALSE) {

  columns <- rlang::enquos(...)

  data %>%
    select(!!! columns) %>%
    transmute(!!sum_col := rowSums(., na.rm = na.rm)) %>%
    bind_cols(data, .)
}

Answer 1

我们可以把...放在最后

rowwise_sum <- function(data, na.rm = FALSE,...) {
  columns <- rlang::enquos(...)
  data %>%
     select(!!!columns) %>%
     rowSums(na.rm = na.rm)
}

cars %>% 
     mutate(sum = rowwise_sum(., na.rm = TRUE, speed, dist))
# A tibble: 50 x 3
#   speed  dist   sum
#   <dbl> <dbl> <dbl>
# 1     4     2     6
# 2     4    10    14
# 3     7     4    11
# 4     7    22    29
# 5     8    16    24
# 6     9    10    19
# 7    10    18    28
# 8    10    26    36
# 9    10    34    44
#10    11    17    28
# ... with 40 more rows

---

它也可以在不改变... 的位置的情况下工作（尽管通常建议这样做）。这里的主要问题是data（即.）未在mutate 的参数列表中指定。

---

在函数中创建整个流程而不是做一部分会更容易

rowwise_sum2 <- function(data, na.rm = FALSE, ...) {
  columns <- rlang::enquos(...)
  data %>%
      select(!!! columns) %>%
      transmute(sum = rowSums(., na.rm = TRUE)) %>%
      bind_cols(data, .)

}

rowwise_sum2(cars, na.rm = TRUE, speed, dist)
# A tibble: 50 x 3
#   speed  dist   sum
#   <dbl> <dbl> <dbl>
# 1     4     2     6
# 2     4    10    14
# 3     7     4    11
# 4     7    22    29
# 5     8    16    24
# 6     9    10    19
# 7    10    18    28
# 8    10    26    36
# 9    10    34    44
#10    11    17    28