Skip to content
/ foreman Public

Unpacking, Interrogating and Subsetting R packages

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE
MIT
LICENSE.md
Notifications You must be signed in to change notification settings

yonicd/foreman

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

foreman

Lifecycle: experimental

The goals of foreman are to:

  • Unpack a package’s functions to interrogate relationships of the functions within it.
  • Isolate function scripts within a package (including the documentation for local paths)
  • Consolidate a subset of self contained functions in a file(s) to allow for focused learning on a specific package functionality.

Given these goals it is important to state that this package is not meant to replace any parent package.

The package supports both local packages and compiled libraries.

Installation

remotes::install_github("yonicd/foreman")

Local Packages

This example will use a local fork of purrr.

library(foreman)
library(ggraph)
library(igraph)

Unpacking

Unpack a pacakge into a list

x <- unpack(path = '../forks/purrr/R', warn = FALSE)

Click the triangle to view the contents found in arrays.R

details::details(lapply(x$arrays.R,get_text),summary = 'arrays.R')
arrays.R
$array_branch
 [1] "#' Coerce array to list"                                              
 [2] "#'"                                                                   
 [3] "#' `array_branch()` and `array_tree()` enable arrays to be"           
 [4] "#' used with purrr's functionals by turning them into lists. The"     
 [5] "#' details of the coercion are controlled by the `margin`"            
 [6] "#' argument. `array_tree()` creates an hierarchical list (a tree)"    
 [7] "#' that has as many levels as dimensions specified in `margin`,"      
 [8] "#' while `array_branch()` creates a flat list (by analogy, a"         
 [9] "#' branch) along all mentioned dimensions."                           
[10] "#'"                                                                   
[11] "#' When no margin is specified, all dimensions are used by"           
[12] "#' default. When `margin` is a numeric vector of length zero, the"    
[13] "#' whole array is wrapped in a list."                                 
[14] "#' @param array An array to coerce into a list."                      
[15] "#' @param margin A numeric vector indicating the positions of the"    
[16] "#'   indices to be to be enlisted. If `NULL`, a full margin is"       
[17] "#'   used. If `numeric(0)`, the array as a whole is wrapped in a"     
[18] "#'   list."                                                           
[19] "#' @name array-coercion"                                              
[20] "#' @export"                                                           
[21] "#' @examples"                                                         
[22] "#' # We create an array with 3 dimensions"                            
[23] "#' x <- array(1:12, c(2, 2, 3))"                                      
[24] "#'"                                                                   
[25] "#' # A full margin for such an array would be the vector 1:3. This is"
[26] "#' # the default if you don't specify a margin"                       
[27] "#'"                                                                   
[28] "#' # Creating a branch along the full margin is equivalent to"        
[29] "#' # as.list(array) and produces a list of size length(x):"           
[30] "#' array_branch(x) %>% str()"                                         
[31] "#'"                                                                   
[32] "#' # A branch along the first dimension yields a list of length 2"    
[33] "#' # with each element containing a 2x3 array:"                       
[34] "#' array_branch(x, 1) %>% str()"                                      
[35] "#'"                                                                   
[36] "#' # A branch along the first and third dimensions yields a list of"  
[37] "#' # length 2x3 whose elements contain a vector of length 2:"         
[38] "#' array_branch(x, c(1, 3)) %>% str()"                                
[39] "#'"                                                                   
[40] "#' # Creating a tree from the full margin creates a list of lists of" 
[41] "#' # lists:"                                                          
[42] "#' array_tree(x) %>% str()"                                           
[43] "#'"                                                                   
[44] "#' # The ordering and the depth of the tree are controlled by the"    
[45] "#' # margin argument:"                                                
[46] "#' array_tree(x, c(3, 1)) %>% str()"                                  
[47] "array_branch <- function(array, margin = NULL) {"                     
[48] "  dims <- dim(array) %||% length(array)"                              
[49] "  margin <- margin %||% seq_along(dims)"                              
[50] ""                                                                     
[51] "  if (length(margin) == 0) {"                                         
[52] "    list(array)"                                                      
[53] "  } else if (is.null(dim(array))) {"                                  
[54] "    if (!identical(as.integer(margin), 1L)) {"                        
[55] "      abort(sprintf("                                                 
[56] "        \"`margin` must be `NULL` or `1` with 1D arrays, not `%s`\"," 
[57] "        toString(margin)"                                             
[58] "      ))"                                                             
[59] "    }"                                                                
[60] "    as.list(array)"                                                   
[61] "  } else {"                                                           
[62] "    flatten(apply(array, margin, list))"                              
[63] "  }"                                                                  
[64] "}"                                                                    

$array_tree
 [1] "#' @rdname array-coercion"                                                     
 [2] "#' @export"                                                                    
 [3] "array_tree <- function(array, margin = NULL) {"                                
 [4] "  dims <- dim(array) %||% length(array)"                                       
 [5] "  margin <- margin %||% seq_along(dims)"                                       
 [6] ""                                                                              
 [7] "  if (length(margin) > 1) {"                                                   
 [8] "    new_margin <- ifelse(margin[-1] > margin[[1]], margin[-1] - 1, margin[-1])"
 [9] "    apply(array, margin[[1]], array_tree, new_margin)"                         
[10] "  } else {"                                                                    
[11] "    array_branch(array, margin)"                                               
[12] "  }"                                                                           
[13] "}"                                                                             

Relationships

x_rel <- relationship(x)

Relationships contained in arrays.R

x_rel$arrays.R
#> $array_branch
#> [1] "flatten"
#> 
#> $array_tree
#> [1] "array_branch"

Functions that compose calls

relationship(x,parent = 'compose')
#> $compose.R
#> $compose.R$compose
#> [1] "compose" "map"     "fn"     
#> 
#> 
#> attr(,"class")
#> [1] "relationship" "list"

Functions who call flatten

relationship(x,child = 'flatten')
#> $arrays.R
#> $arrays.R$array_branch
#> [1] "flatten"
#> 
#> 
#> $lmap.R
#> $lmap.R$lmap_at
#> [1] "flatten"
#> 
#> 
#> $splice.R
#> $splice.R$splice_if
#> [1] "flatten"
#> 
#> 
#> attr(,"class")
#> [1] "relationship" "list"

Convert relationships to a data.frame

x_rel_df <- as.data.frame(x_rel)

Click the triangle to view the data.frame

details::details(x_rel_df,summary = 'Relatives')
Relatives
                           child                 parent               file
1                        flatten           array_branch           arrays.R
2                      as_mapper            as_function        as_mapper.R
3                   stop_defunct            as_function        as_mapper.R
4                     paste_line            as_function        as_mapper.R
5                         coerce             coerce_lgl           coerce.R
6                   can_simplify              as_vector         coercion.R
7                warn_deprecated signal_soft_deprecated compat-lifecycle.R
8                        compose                compose          compose.R
9                            map                compose          compose.R
10                            fn                compose          compose.R
11                     as_vector                lift_vl      composition.R
12               what_bad_object          stop_bad_type       conditions.R
13              friendly_type_of          stop_bad_type       conditions.R
14                     as_mapper                  cross            cross.R
15 as_predicate_friendly_type_of                  cross            cross.R
16                       compact                  cross            cross.R
17                       is_bool                  cross            cross.R
18                       map_int              vec_depth            depth.R
19                  as_predicate                 detect           detect.R
20                         index                 detect           detect.R
21                  as_predicate                  every       every-some.R
22                  detect_index             head_while        head-tail.R
23                        negate             head_while        head-tail.R
24                     as_mapper                   imap             imap.R
25                     vec_index                   imap             imap.R
26                          map2                   imap             imap.R
27                         probe                   keep             keep.R
28                  list_recurse            list_modify      list-modify.R
29                       lmap_at                   lmap             lmap.R
30                     as_mapper                    map              map.R
31                     as_mapper                   map2        map2-pmap.R
32                     as_mapper         modify.default           modify.R
33                     as_mapper                 negate           negate.R
34                     as_mapper                 safely           output.R
35                 capture_error                 safely           output.R
36        signal_soft_deprecated                partial          partial.R
37                  stop_defunct                partial          partial.R
38                           map                partial          partial.R
39                    paste_line                partial          partial.R
40              friendly_type_of                partial          partial.R
41                     assign_in              `pluck<-`            pluck.R
42                     as_mapper            insistently             rate.R
43                 stop_bad_type            insistently             rate.R
44                 capture_error            insistently             rate.R
45                  rate_backoff            insistently             rate.R
46                       is_rate            insistently             rate.R
47                    rate_sleep            insistently             rate.R
48                    rate_reset            insistently             rate.R
49                             f            insistently             rate.R
50                   reduce_impl                 reduce           reduce.R
51                     eval_dots                  rerun            rerun.R
52                     has_names                  rerun            rerun.R
53                          map2             invoke_map   retired-invoke.R
54            as_invoke_function             invoke_map   retired-invoke.R
55                     splice_if                 splice           splice.R
56                  check_tibble    maybe_as_data_frame            utils.R

Plotting the relationships

graph <- igraph::graph_from_data_frame(x_rel_df,directed = TRUE)
igraph::V(graph)$parents <- names(igraph::V(graph))
ggraph(graph) +
  geom_edge_link(
    aes(colour = file),
    arrow = grid::arrow(length = unit(0.05, "inches"))) +
  geom_node_text(aes(label = parents),size = 3) +
  labs(title = 'purrr function map', colour = 'Exported') + 
  ggplot2::theme(legend.position   = 'bottom')
#> Using `nicely` as default layout

Subset

Subsetting Package Functions

sub_x <- subset(x,'compose')

Click the triangle to view the contents found in the subset containing compose and the functions the it calls.

details::details(lapply(sub_x,get_text),summary = 'Package subset')
Package subset
$compose.R
 [1] "#' Compose multiple functions"                                           
 [2] "#'"                                                                      
 [3] "#' @param ... Functions to apply in order (from right to left by"        
 [4] "#'   default). Formulas are converted to functions in the usual way."    
 [5] "#'"                                                                      
 [6] "#'   These dots support [tidy dots][rlang::list2] features. In"          
 [7] "#'   particular, if your functions are stored in a list, you can"        
 [8] "#'   splice that in with `!!!`."                                         
 [9] "#' @param .dir If `\"backward\"` (the default), the functions are called"
[10] "#'   in the reverse order, from right to left, as is conventional in"    
[11] "#'   mathematics. If `\"forward\"`, they are called from left to right." 
[12] "#' @return A function"                                                   
[13] "#' @export"                                                              
[14] "#' @examples"                                                            
[15] "#' not_null <- compose(`!`, is.null)"                                    
[16] "#' not_null(4)"                                                          
[17] "#' not_null(NULL)"                                                       
[18] "#'"                                                                      
[19] "#' add1 <- function(x) x + 1"                                            
[20] "#' compose(add1, add1)(8)"                                               
[21] "#'"                                                                      
[22] "#' # You can use the formula shortcut for functions:"                    
[23] "#' fn <- compose(~ paste(.x, \"foo\"), ~ paste(.x, \"bar\"))"            
[24] "#' fn(\"input\")"                                                        
[25] "#'"                                                                      
[26] "#' # Lists of functions can be spliced with !!!"                         
[27] "#' fns <- list("                                                         
[28] "#'   function(x) paste(x, \"foo\"),"                                     
[29] "#'   ~ paste(.x, \"bar\")"                                               
[30] "#' )"                                                                    
[31] "#' fn <- compose(!!!fns)"                                                
[32] "#' fn(\"input\")"                                                        
[33] "compose <- function(..., .dir = c(\"backward\", \"forward\")) {"         
[34] "  .dir <- arg_match(.dir, c(\"backward\", \"forward\"))"                 
[35] ""                                                                        
[36] "  fns <- map(list2(...), rlang::as_closure, env = caller_env())"         
[37] "  if (!length(fns)) {"                                                   
[38] "    # Return the identity function"                                      
[39] "    return(compose(function(x, ...) x))"                                 
[40] "  }"                                                                     
[41] ""                                                                        
[42] "  if (.dir == \"backward\") {"                                           
[43] "    n <- length(fns)"                                                    
[44] "    first_fn <- fns[[n]]"                                                
[45] "    fns <- rev(fns[-n])"                                                 
[46] "  } else {"                                                              
[47] "    first_fn <- fns[[1]]"                                                
[48] "    fns <- fns[-1]"                                                      
[49] "  }"                                                                     
[50] ""                                                                        
[51] "  body <- expr({"                                                        
[52] "    out <- !!fn_body(first_fn)"                                          
[53] ""                                                                        
[54] "    fns <- !!fns"                                                        
[55] "    for (fn in fns) {"                                                   
[56] "      out <- fn(out)"                                                    
[57] "    }"                                                                   
[58] ""                                                                        
[59] "    out"                                                                 
[60] "  })"                                                                    
[61] ""                                                                        
[62] "  structure("                                                            
[63] "    new_function(formals(first_fn), body, fn_env(first_fn)),"            
[64] "    class = c(\"purrr_function_compose\", \"function\"),"                
[65] "    first_fn = first_fn,"                                                
[66] "    fns = fns"                                                           
[67] "  )"                                                                     
[68] "}"                                                                       

$map.R
  [1] "#' Apply a function to each element of a vector"                              
  [2] "#'"                                                                           
  [3] "#' @description"                                                              
  [4] "#'"                                                                           
  [5] "#' The map functions transform their input by applying a function to"         
  [6] "#' each element and returning a vector the same length as the input."         
  [7] "#'"                                                                           
  [8] "#' * `map()`, `map_if()` and `map_at()` always return a list. See the"        
  [9] "#'   [modify()] family for versions that return an object of the same"        
 [10] "#'   type as the input."                                                      
 [11] "#'"                                                                           
 [12] "#'   The `_if` and `_at` variants take a predicate function `.p` that"        
 [13] "#'   determines which elements of `.x` are transformed with `.f`."            
 [14] "#'"                                                                           
 [15] "#' * `map_lgl()`, `map_int()`, `map_dbl()` and `map_chr()` each return"       
 [16] "#'    an atomic vector of the indicated type (or die trying)."                
 [17] "#'"                                                                           
 [18] "#'    The return value of `.f` must be of length one for each element of"     
 [19] "#'    `.x`. If `.f` uses an extractor function shortcut, `.default`"          
 [20] "#'    can be specified to handle values that are absent or empty.  See"       
 [21] "#'    [as_mapper()] for more on `.default`."                                  
 [22] "#'"                                                                           
 [23] "#' * `map_dfr()` and `map_dfc()` return data frames created by"               
 [24] "#'   row-binding and column-binding respectively. They require dplyr"         
 [25] "#'   to be installed."                                                        
 [26] "#'"                                                                           
 [27] "#' * `walk()` calls `.f` for its side-effect and returns the input `.x`."     
 [28] "#'"                                                                           
 [29] "#' @inheritParams as_mapper"                                                  
 [30] "#' @param .x A list or atomic vector."                                        
 [31] "#' @param .p A single predicate function, a formula describing such a"        
 [32] "#'   predicate function, or a logical vector of the same length as `.x`."     
 [33] "#'   Alternatively, if the elements of `.x` are themselves lists of"          
 [34] "#'   objects, a string indicating the name of a logical element in the"       
 [35] "#'   inner lists. Only those elements where `.p` evaluates to"                
 [36] "#'   `TRUE` will be modified."                                                
 [37] "#' @param .at A character vector of names, positive numeric vector of"        
 [38] "#'   positions to include, or a negative numeric vector of positions to"      
 [39] "#'   exlude. Only those elements corresponding to `.at` will be modified."    
 [40] "#' @param ... Additional arguments passed on to the mapped function."         
 [41] "#' @return All functions return a vector the same length as `.x`."            
 [42] "#'"                                                                           
 [43] "#'   `map()` returns a list, `map_lgl()` a logical vector, `map_int()` an"    
 [44] "#'   integer vector, `map_dbl()` a double vector, and `map_chr()` a character"
 [45] "#'   vector. The output of `.f` will be automatically typed upwards,"         
 [46] "#'   e.g. logical -> integer -> double -> character."                         
 [47] "#'"                                                                           
 [48] "#'   If `.x` has `names()`, the return value preserves those names."          
 [49] "#'"                                                                           
 [50] "#'   `walk()` returns the input `.x` (invisibly). This makes it easy to"      
 [51] "#'   use in pipe."                                                            
 [52] "#' @export"                                                                   
 [53] "#' @family map variants"                                                      
 [54] "#' @examples"                                                                 
 [55] "#' 1:10 %>%"                                                                  
 [56] "#'   map(rnorm, n = 10) %>%"                                                  
 [57] "#'   map_dbl(mean)"                                                           
 [58] "#'"                                                                           
 [59] "#' # Or use an anonymous function"                                            
 [60] "#' 1:10 %>%"                                                                  
 [61] "#'   map(function(x) rnorm(10, x))"                                           
 [62] "#'"                                                                           
 [63] "#' # Or a formula"                                                            
 [64] "#' 1:10 %>%"                                                                  
 [65] "#'   map(~ rnorm(10, .x))"                                                    
 [66] "#'"                                                                           
 [67] "#' # The names of the input are preserved in the output:"                     
 [68] "#' list(foo = 1, bar = 2) %>% map(`+`, 10)"                                   
 [69] "#'"                                                                           
 [70] "#' # Using set_names() with character vectors is handy to keep track"         
 [71] "#' # of the original inputs:"                                                 
 [72] "#' set_names(c(\"foo\", \"bar\")) %>% map_chr(paste0, \":suffix\")"           
 [73] "#'"                                                                           
 [74] "#' # Extract by name or position"                                             
 [75] "#' # .default specifies value for elements that are missing or NULL"          
 [76] "#' l1 <- list(list(a = 1L), list(a = NULL, b = 2L), list(b = 3L))"            
 [77] "#' l1 %>% map(\"a\", .default = \"???\")"                                     
 [78] "#' l1 %>% map_int(\"b\", .default = NA)"                                      
 [79] "#' l1 %>% map_int(2, .default = NA)"                                          
 [80] "#'"                                                                           
 [81] "#' # Supply multiple values to index deeply into a list"                      
 [82] "#' l2 <- list("                                                               
 [83] "#'   list(num = 1:3,     letters[1:3]),"                                      
 [84] "#'   list(num = 101:103, letters[4:6]),"                                      
 [85] "#'   list()"                                                                  
 [86] "#' )"                                                                         
 [87] "#' l2 %>% map(c(2, 2))"                                                       
 [88] "#'"                                                                           
 [89] "#' # Use a list to build an extractor that mixes numeric indices and names,"  
 [90] "#' # and .default to provide a default value if the element does not exist"   
 [91] "#' l2 %>% map(list(\"num\", 3))"                                              
 [92] "#' l2 %>% map_int(list(\"num\", 3), .default = NA)"                           
 [93] "#'"                                                                           
 [94] "#'"                                                                           
 [95] "#' # Use a predicate function to decide whether to map a function:"           
 [96] "#' map_if(iris, is.factor, as.character)"                                     
 [97] "#'"                                                                           
 [98] "#' # Specify an alternative with the `.else` argument:"                       
 [99] "#' map_if(iris, is.factor, as.character, .else = as.integer)"                 
[100] "#'"                                                                           
[101] "#' # A more realistic example: split a data frame into pieces, fit a"         
[102] "#' # model to each piece, summarise and extract R^2"                          
[103] "#' mtcars %>%"                                                                
[104] "#'   split(.$cyl) %>%"                                                        
[105] "#'   map(~ lm(mpg ~ wt, data = .x)) %>%"                                      
[106] "#'   map(summary) %>%"                                                        
[107] "#'   map_dbl(\"r.squared\")"                                                  
[108] "#'"                                                                           
[109] "#' # Use map_lgl(), map_dbl(), etc to reduce to a vector."                    
[110] "#' # * list"                                                                  
[111] "#' mtcars %>% map(sum)"                                                       
[112] "#' # * vector"                                                                
[113] "#' mtcars %>% map_dbl(sum)"                                                   
[114] "#'"                                                                           
[115] "#' # If each element of the output is a data frame, use"                      
[116] "#' # map_dfr to row-bind them together:"                                      
[117] "#' mtcars %>%"                                                                
[118] "#'   split(.$cyl) %>%"                                                        
[119] "#'   map(~ lm(mpg ~ wt, data = .x)) %>%"                                      
[120] "#'   map_dfr(~ as.data.frame(t(as.matrix(coef(.)))))"                         
[121] "#' # (if you also want to preserve the variable names see"                    
[122] "#' # the broom package)"                                                      
[123] "#'"                                                                           
[124] "#' # Use `map_depth()` to recursively traverse nested vectors and map"        
[125] "#' # a function at a certain depth:"                                          
[126] "#' x <- list(a = list(foo = 1:2, bar = 3:4), b = list(baz = 5:6))"            
[127] "#' str(x)"                                                                    
[128] "#' map_depth(x, 2, paste, collapse = \"/\")"                                  
[129] "#'"                                                                           
[130] "#' # Equivalent to:"                                                          
[131] "#' map(x, map, paste, collapse = \"/\")"                                      
[132] "map <- function(.x, .f, ...) {"                                               
[133] "  .f <- as_mapper(.f, ...)"                                                   
[134] "  .Call(map_impl, environment(), \".x\", \".f\", \"list\")"                   
[135] "}"                                                                            

$reduce.R
[1] "    fn <- function(x, y, ...) .f(y, x, ...)"

Consolidation

Consolidating Subsetted Functions into a File

pack_path <- repack(sub_x)
#> Functions packed to /var/folders/kx/t4h_mm1910sb7vhm_gnfnx2c0000gn/T//Rtmpgg7cm0/foreman/unpacked.R

Click the triangle to view the contents found in the file containing the consolidated functions.

details::details(file.path(pack_path,'unpacked.R'),summary = 'Consolidated Script')
Consolidated Script
#Generated by foreman:

#' Compose multiple functions
#'
#' @param ... Functions to apply in order (from right to left by
#'   default). Formulas are converted to functions in the usual way.
#'
#'   These dots support [tidy dots][rlang::list2] features. In
#'   particular, if your functions are stored in a list, you can
#'   splice that in with `!!!`.
#' @param .dir If `"backward"` (the default), the functions are called
#'   in the reverse order, from right to left, as is conventional in
#'   mathematics. If `"forward"`, they are called from left to right.
#' @return A function
#' @export
#' @examples
#' not_null <- compose(`!`, is.null)
#' not_null(4)
#' not_null(NULL)
#'
#' add1 <- function(x) x + 1
#' compose(add1, add1)(8)
#'
#' # You can use the formula shortcut for functions:
#' fn <- compose(~ paste(.x, "foo"), ~ paste(.x, "bar"))
#' fn("input")
#'
#' # Lists of functions can be spliced with !!!
#' fns <- list(
#'   function(x) paste(x, "foo"),
#'   ~ paste(.x, "bar")
#' )
#' fn <- compose(!!!fns)
#' fn("input")
compose <- function(..., .dir = c("backward", "forward")) {
  .dir <- arg_match(.dir, c("backward", "forward"))

  fns <- map(list2(...), rlang::as_closure, env = caller_env())
  if (!length(fns)) {
    # Return the identity function
    return(compose(function(x, ...) x))
  }

  if (.dir == "backward") {
    n <- length(fns)
    first_fn <- fns[[n]]
    fns <- rev(fns[-n])
  } else {
    first_fn <- fns[[1]]
    fns <- fns[-1]
  }

  body <- expr({
    out <- !!fn_body(first_fn)

    fns <- !!fns
    for (fn in fns) {
      out <- fn(out)
    }

    out
  })

  structure(
    new_function(formals(first_fn), body, fn_env(first_fn)),
    class = c("purrr_function_compose", "function"),
    first_fn = first_fn,
    fns = fns
  )
}
#' Apply a function to each element of a vector
#'
#' @description
#'
#' The map functions transform their input by applying a function to
#' each element and returning a vector the same length as the input.
#'
#' * `map()`, `map_if()` and `map_at()` always return a list. See the
#'   [modify()] family for versions that return an object of the same
#'   type as the input.
#'
#'   The `_if` and `_at` variants take a predicate function `.p` that
#'   determines which elements of `.x` are transformed with `.f`.
#'
#' * `map_lgl()`, `map_int()`, `map_dbl()` and `map_chr()` each return
#'    an atomic vector of the indicated type (or die trying).
#'
#'    The return value of `.f` must be of length one for each element of
#'    `.x`. If `.f` uses an extractor function shortcut, `.default`
#'    can be specified to handle values that are absent or empty.  See
#'    [as_mapper()] for more on `.default`.
#'
#' * `map_dfr()` and `map_dfc()` return data frames created by
#'   row-binding and column-binding respectively. They require dplyr
#'   to be installed.
#'
#' * `walk()` calls `.f` for its side-effect and returns the input `.x`.
#'
#' @inheritParams as_mapper
#' @param .x A list or atomic vector.
#' @param .p A single predicate function, a formula describing such a
#'   predicate function, or a logical vector of the same length as `.x`.
#'   Alternatively, if the elements of `.x` are themselves lists of
#'   objects, a string indicating the name of a logical element in the
#'   inner lists. Only those elements where `.p` evaluates to
#'   `TRUE` will be modified.
#' @param .at A character vector of names, positive numeric vector of
#'   positions to include, or a negative numeric vector of positions to
#'   exlude. Only those elements corresponding to `.at` will be modified.
#' @param ... Additional arguments passed on to the mapped function.
#' @return All functions return a vector the same length as `.x`.
#'
#'   `map()` returns a list, `map_lgl()` a logical vector, `map_int()` an
#'   integer vector, `map_dbl()` a double vector, and `map_chr()` a character
#'   vector. The output of `.f` will be automatically typed upwards,
#'   e.g. logical -> integer -> double -> character.
#'
#'   If `.x` has `names()`, the return value preserves those names.
#'
#'   `walk()` returns the input `.x` (invisibly). This makes it easy to
#'   use in pipe.
#' @export
#' @family map variants
#' @examples
#' 1:10 %>%
#'   map(rnorm, n = 10) %>%
#'   map_dbl(mean)
#'
#' # Or use an anonymous function
#' 1:10 %>%
#'   map(function(x) rnorm(10, x))
#'
#' # Or a formula
#' 1:10 %>%
#'   map(~ rnorm(10, .x))
#'
#' # The names of the input are preserved in the output:
#' list(foo = 1, bar = 2) %>% map(`+`, 10)
#'
#' # Using set_names() with character vectors is handy to keep track
#' # of the original inputs:
#' set_names(c("foo", "bar")) %>% map_chr(paste0, ":suffix")
#'
#' # Extract by name or position
#' # .default specifies value for elements that are missing or NULL
#' l1 <- list(list(a = 1L), list(a = NULL, b = 2L), list(b = 3L))
#' l1 %>% map("a", .default = "???")
#' l1 %>% map_int("b", .default = NA)
#' l1 %>% map_int(2, .default = NA)
#'
#' # Supply multiple values to index deeply into a list
#' l2 <- list(
#'   list(num = 1:3,     letters[1:3]),
#'   list(num = 101:103, letters[4:6]),
#'   list()
#' )
#' l2 %>% map(c(2, 2))
#'
#' # Use a list to build an extractor that mixes numeric indices and names,
#' # and .default to provide a default value if the element does not exist
#' l2 %>% map(list("num", 3))
#' l2 %>% map_int(list("num", 3), .default = NA)
#'
#'
#' # Use a predicate function to decide whether to map a function:
#' map_if(iris, is.factor, as.character)
#'
#' # Specify an alternative with the `.else` argument:
#' map_if(iris, is.factor, as.character, .else = as.integer)
#'
#' # A more realistic example: split a data frame into pieces, fit a
#' # model to each piece, summarise and extract R^2
#' mtcars %>%
#'   split(.$cyl) %>%
#'   map(~ lm(mpg ~ wt, data = .x)) %>%
#'   map(summary) %>%
#'   map_dbl("r.squared")
#'
#' # Use map_lgl(), map_dbl(), etc to reduce to a vector.
#' # * list
#' mtcars %>% map(sum)
#' # * vector
#' mtcars %>% map_dbl(sum)
#'
#' # If each element of the output is a data frame, use
#' # map_dfr to row-bind them together:
#' mtcars %>%
#'   split(.$cyl) %>%
#'   map(~ lm(mpg ~ wt, data = .x)) %>%
#'   map_dfr(~ as.data.frame(t(as.matrix(coef(.)))))
#' # (if you also want to preserve the variable names see
#' # the broom package)
#'
#' # Use `map_depth()` to recursively traverse nested vectors and map
#' # a function at a certain depth:
#' x <- list(a = list(foo = 1:2, bar = 3:4), b = list(baz = 5:6))
#' str(x)
#' map_depth(x, 2, paste, collapse = "/")
#'
#' # Equivalent to:
#' map(x, map, paste, collapse = "/")
map <- function(.x, .f, ...) {
  .f <- as_mapper(.f, ...)
  .Call(map_impl, environment(), ".x", ".f", "list")
}
    fn <- function(x, y, ...) .f(y, x, ...)

Compiled Library

This example will use the installed library future.

Using foreman with an compiled libraries is also simple

library(future)
#> 
#> Attaching package: 'future'
#> The following objects are masked from 'package:igraph':
#> 
#>     %->%, %<-%

unpacked_future <- unpack(ns = 'future')%>%
  relationship()%>%
  as.data.frame()
graph <- igraph::graph_from_data_frame(unpacked_future,directed = TRUE)
igraph::V(graph)$parents <- names(igraph::V(graph))
igraph::V(graph)$exported <- names(igraph::V(graph))%in%ls('package:future')
ggraph(graph) +
  geom_edge_link(
    arrow = grid::arrow(length = unit(0.05, "inches")),alpha = 0.05) +
  geom_node_text(aes(colour = exported,label = parents),size = 2) +
  labs(title = 'future function map', colour = 'Exported') + 
  ggplot2::theme(legend.position   = 'bottom')
#> Using `nicely` as default layout

About

Unpacking, Interrogating and Subsetting R packages

Resources

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE
MIT
LICENSE.md

Stars

Watchers

Forks

Packages

No packages published

Languages