dplyr::do 内函数的方法调度

Question

如何为 dplyr::do 中的函数实现方法分派？

我通读了 GitHub 个问题 #719, #3558 and #3429，其中包含有关如何为 dplyr 动词创建方法的有用信息，但没有特别适用于 dplyr::do - 这有点像 "special"，因为调度不仅需要发生在 dplyr:do 本身，而且还发生在 dplyr::do 内部调用的函数（或者至少我是这样的）之后）

这是我尝试过的：

预赛

library(dplyr)
#> 
#> Attache Paket: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

# Example data ------------------------------------------------------------

df <- tibble::tibble(
  id = c(rep("A", 5), rep("B", 5)),
  x = 1:10
)

df_custom <- df
class(df_custom) <- c("tbl_df_custom", class(df_custom))

# Reclass function --------------------------------------------------------

reclass <- function(x, result) {
  UseMethod('reclass')
}

reclass.default <- function(x, result) {
  class(result) <- unique(c(class(x)[[1]], class(result)))
  attr(result, class(x)[[1]]) <- attr(x, class(x)[[1]])
  result
}

第 1 步：尝试为 dplyr 动词定义一个方法

# Custom method for summarize ---------------------------------------------

summarise.tbl_df_custom <- function (.data, ...) {
  message("Custom method for `summarise`")
  result <- NextMethod("summarise")
  ret <- reclass(.data, result)
  print(class(ret))
  ret
}

ret <- df_custom %>%
  summarise(y = mean(x))
#> Custom method for `summarise`
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"
ret %>% class()
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"

第 2 步：尝试为另一个 dplyr 动词定义一个方法来测试更长的管道

# Custom method for group_by ----------------------------------------------

group_by.tbl_df_custom <- function (.data, ..., add = FALSE) {
  message("Custom method for `group_by`")
  result <- NextMethod("group_by")
  ret <- reclass(.data, result)
  print(class(ret))
  ret
}

ret <- df_custom %>%
  group_by(id) %>%
  summarise(y = mean(x))
#> Custom method for `group_by`
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"
#> Custom method for `summarise`
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"
ret %>% class()
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"

第 3 步：对 `do`

进行同样的尝试

# Custom method for do ----------------------------------------------------

do.tbl_df_custom <- function (.data, ...) {
  message("custom method for `do`")
  result <- NextMethod("do")
  ret <- reclass(.data, result)
  print(class(ret))
  ret
}

foo <- function(df) {
  UseMethod("foo")
}

foo.default <- function(df) {
  message("Default method for `foo`")
  df %>%
    summarise(y = mean(x))
}

foo.tbl_df_custom <- function(df) {
  message("Custom method for `foo`")
  df %>%
    summarise(y = mean(x) * 100)
}

ret <- df_custom %>%
  group_by(id) %>%
  do(foo(.))
#> Custom method for `group_by`
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"
#> custom method for `do`
#> Default method for `foo`
#> Default method for `foo`
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"
ret
#> # A tibble: 2 x 2
#> # Groups:   id [2]
#>   id        y
#>   <chr> <dbl>
#> 1 A         3
#> 2 B         8
ret %>% class()
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"

虽然乍一看这看起来不错，但问题是 default 而不是 foo 的 custom 方法被调用。

由 reprex package (v0.2.1)

创建于 2019-01-08

Answer 1

所以这个问题与有关。我能够通过定义 3 个新函数来解决它：ungroup.tbl_df_custom、class 构造函数和 [.tbl_df_custom.

ungroup.tbl_df_custom <- function (.data, ...) {
  message("custom method for `ungroup`")
  result <- NextMethod("ungroup")
  ret <- reclass(.data, result)
  ret
}


new_custom <- function(x, ...) {

  structure(x, class = c("tbl_df_custom", class(x)))
}

`[.tbl_df_custom` <- function(x, ...) {
  new_custom(NextMethod())
}



df_custom2 <- new_custom(df)


df_custom2 %>%
  group_by(id) %>%
  do(foo(.))

Custom method for `group_by`
[1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"           "data.frame"   
custom method for `do`
custom method for `ungroup`
Custom method for `foo`
Custom method for `summarise`
[1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"   
Custom method for `foo`
Custom method for `summarise`
[1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"   
[1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"           "data.frame"   
custom method for `ungroup`
# A tibble: 2 x 2
# Groups:   id [2]
  id        y
  <chr> <dbl>
1 A       300
2 B       800

Answer 2

纯粹为了拥有一个完整且自包含的示例以及我的特定示例从头到尾的所有代码，我还将post在此处提供自己的答案。

有几件事要强调：

除了我为 group_by() 定制的方法外，我可以将 reclass() 换成更好的 vctrs::vec_restore()，它也恰好有一个 data.frame ] 方法（参见 library(vctrs); sloop::s3_methods_generic("vec_restore")）。

您可以在 S3 inheritance of Advanced R as well as as the S3 vectors article on https://vctrs.r-lib.org/
章中找到有关 vctrs::vec_restore() 的更多信息
如果 vctrs::vec_restore() 中有类似 combine 的参数，让它考虑通过调用group_by() 的默认方法，但这是另一个故事（我为此提出了一个好奇的 GitHub issue）。

目前，由于 vctrs::vec_restore() 的实施方式（请参阅下面的 "Testing things out"），我们的自定义 class 信息将被删除。
GitHub 个我发现非常有用的问题：#3429 and especially #3923

代码

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

# Constructor for tbl_df_custom class -------------------------------------

new_df_custom <- function(x = tibble()) {
  stopifnot(tibble::is_tibble(x))
  structure(x, class = c("tbl_df_custom", class(x)))
}

# Example data ------------------------------------------------------------

df_custom <- new_df_custom(
  x = tibble::tibble(
    id = c(rep("A", 3), rep("B", 3)),
    x = 1:6
  )
)

df_custom
#> # A tibble: 6 x 2
#>   id        x
#> * <chr> <int>
#> 1 A         1
#> 2 A         2
#> 3 A         3
#> 4 B         4
#> 5 B         5
#> 6 B         6
df_custom %>% class()
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"

# Reclass function for preserving custom class attribute ------------------

reclass <- function(x, to) {
  UseMethod('reclass')
}

reclass.default <- function(x, to) {
  class(x) <- unique(c(class(to)[[1]], class(x)))
  attr(x, class(to)[[1]]) <- attr(to, class(to)[[1]])
  x
}

# Custom method for summarise ---------------------------------------------

summarise.tbl_df_custom <- function (.data, ...) {
  message("Custom method for `summarise`")
  vctrs::vec_restore(NextMethod(), .data)
}

# Custom method for group_by ----------------------------------------------

group_by.tbl_df_custom <- function (.data, ..., add = FALSE, 
  use_vec_restore = FALSE
) {
  message("Custom method for `group_by`")
  retval <- reclass(NextMethod(), .data)
  print(class(retval))
  retval
}

# Custom method for ungroup ----------------------------------------------

ungroup.tbl_df_custom <- function (.data, ...) {
  message("custom method for `ungroup`")
  vctrs::vec_restore(NextMethod(), .data)
}

# Custom method for do ----------------------------------------------------

do.tbl_df_custom <- function (.data, ...) {
  message("custom method for `do`")
  vctrs::vec_restore(NextMethod(), .data)
}

# Custom extraction method ------------------------------------------------

`[.tbl_df_custom` <- function(x, ...) {
  message("custom method for `[`")
  new_df_custom(NextMethod())
}

# Create custom methods for foo -------------------------------------------

foo <- function(df) {
  UseMethod("foo")
}

foo.default <- function(df) {
  message("Default method for `foo`")
  df %>%
    summarise(y = mean(x))
}

foo.tbl_df_custom <- function(df) {
  message("Custom method for `foo`")
  df %>%
    summarise(y = mean(x) * 100)
}

# Testing things out ------------------------------------------------------

retval <- df_custom %>%
  group_by(id) %>%
  do(foo(.))
#> Custom method for `group_by`
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"
#> custom method for `do`
#> custom method for `ungroup`
#> custom method for `[`
#> Custom method for `foo`
#> Custom method for `summarise`
#> custom method for `[`
#> Custom method for `foo`
#> Custom method for `summarise`

retval
#> custom method for `[`
#> custom method for `ungroup`
#> # A tibble: 2 x 2
#> # Groups:   id [2]
#>   id        y
#>   <chr> <dbl>
#> 1 A       200
#> 2 B       500
retval %>% class()
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"

由 reprex package (v0.2.1)

于 2019-01-08 创建

替代`reclass()`：`vctrs::vec_restore()`

# Alternative version for group_by that uses vctrs::vec_restore -----------

group_by.tbl_df_custom <- function (.data, ..., add = FALSE) {
  message("Custom method for `group_by`")
  retval <- vctrs::vec_restore(NextMethod(), .data)
  print(class(retval))
  retval
}

retval <- df_custom %>%
  group_by(id) %>%
  do(foo(.))
#> Custom method for `group_by`
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"
#> custom method for `do`
#> Custom method for `foo`
#> Custom method for `summarise`

retval
#> custom method for `[`
#> # A tibble: 1 x 1
#>       y
#>   <dbl>
#> 1   350
retval %>% class()
#> [1] "tbl_df_custom" "tbl_df"        "tbl"           "data.frame"

由 reprex package (v0.2.1)

于 2019-01-08 创建

如上所述，请注意，当使用 group_by() 的替代版本时，使用 vctrs::vec_restore() 而不是 reclass()，class 属性 grouped_df掉线了。

替代`reclass()`：`vec_restore_inclusive()`

这是一个自己的实现，它试图利用 vctrs::vec_restore() 的工作方式，同时在决定如何执行 "reset" 时也考虑了 to 的属性。可以说，"combine" 或 "align" 是函数的更好名称组件。

vec_restore_inclusive <- function(x, to) {
  UseMethod('vec_restore_inclusive')
}

vec_restore_inclusive.data.frame <- function (x, to) {
  attr_to <- attributes(to)
  attr_x <- attributes(x)
  attr_use <- if (
    length(classes_preserve <- setdiff(attr_to[["class"]], attr_x[["class"]]))
  ) {
    attr_x
  } else {
    attr_to
  }

  attr_use[["names"]] <- attr_x[["names"]]
  attr_use[["row.names"]] <- .set_row_names(vctrs:::df_length(x))
  attr_use[["class"]] <- unique(c(classes_preserve, attr_x[["class"]]))
  attributes(x) <- attr_use
  x
}

group_by.tbl_df_custom <- function (.data, ..., add = FALSE) {
  message("Custom method for `group_by`")
  retval <- vec_restore_inclusive(NextMethod(), .data)
  print(class(retval))
  retval
}

retval <- df_custom %>%
  group_by(id) %>%
  do(foo(.))
#> Custom method for `group_by`
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"
#> custom method for `do`
#> custom method for `ungroup`
#> custom method for `[`
#> Custom method for `foo`
#> Custom method for `summarise`
#> custom method for `[`
#> Custom method for `foo`
#> Custom method for `summarise`

retval
#> custom method for `[`
#> custom method for `ungroup`
#> # A tibble: 2 x 2
#> # Groups:   id [2]
#>   id        y
#>   <chr> <dbl>
#> 1 A       200
#> 2 B       500
retval %>% class()
#> [1] "tbl_df_custom" "grouped_df"    "tbl_df"        "tbl"          
#> [5] "data.frame"

由 reprex package (v0.2.1)

于 2019-01-08 创建

dplyr::do 内函数的方法调度

Method dispatch for functions inside dplyr::do

methods

r

r-s3

dplyr

这是我尝试过的：

预赛

第 1 步：尝试为 dplyr 动词定义一个方法

第 2 步：尝试为另一个 dplyr 动词定义一个方法来测试更长的管道

第 3 步：对 `do`

代码

替代`reclass()`：`vctrs::vec_restore()`

替代`reclass()`：`vec_restore_inclusive()`

dplyr::do 内函数的方法调度

Method dispatch for functions inside dplyr::do

methods

r

r-s3

dplyr

这是我尝试过的：

预赛

第 1 步：尝试为 dplyr 动词定义一个方法

第 2 步：尝试为另一个 dplyr 动词定义一个方法来测试更长的管道

第 3 步：对 do

代码

替代reclass()：vctrs::vec_restore()

替代reclass()：vec_restore_inclusive()

第 3 步：对 `do`

替代`reclass()`：`vctrs::vec_restore()`

替代`reclass()`：`vec_restore_inclusive()`