purrr

Purrr makes your pure functions purr by completing R's functional programming tools with important features from other languages, in the style of the JS packages underscore.js, lodash and lazy.js.

Installation

Get the released version from CRAN:

install.packages("purrr")

Or the development version from github with:

# install.packages("devtools")
devtools::install_github("hadley/purrr")

Examples

The following example uses purrr to solve a fairly realistic problem: split a data frame into pieces, fit a model to each piece, summarise and extract R^2.

library(purrr)

mtcars %>%
  split(.$cyl) %>%
  map(~ lm(mpg ~ wt, data = .)) %>%
  map(summary) %>%
  map_dbl("r.squared")

Note the three types of input to map(): a function, a formula (converted to an anonymous function), or a string (used to extract named components).

The following more complicated example shows how you might generate 100 random test-training splits, fit a model to each training split then evaluate based on the test split:

library(dplyr)
random_group <- function(n, probs) {
  probs <- probs / sum(probs)
  g <- findInterval(seq(0, 1, length = n), c(0, cumsum(probs)),
    rightmost.closed = TRUE)
  names(probs)[sample(g)]
}
partition <- function(df, n, probs) {
  replicate(n, split(df, random_group(nrow(df), probs)), FALSE) %>%
    transpose() %>%
    as_data_frame()
}

msd <- function(x, y) sqrt(mean((x - y) ^ 2))

# Generate 100 random test-training splits
boot <- partition(mtcars, 100, c(training = 0.8, test = 0.2))
boot

boot <- boot %>% mutate(
  # Fit the models
  models = map(training, ~ lm(mpg ~ wt, data = .)),
  # Make predictions on test data
  preds = map2(models, test, predict),
  diffs = map2(preds, test %>% map("mpg"), msd)
)

# Evaluate mean-squared difference between predicted and actual
mean(unlist(boot$diffs))

API

Transformation

• Apply a function to each element: map() returns a list; map_lgl()/map_int()/map_dbl()/map_chr() return a vector; walk() invisibly returns original list, calling the function for its side effects; map2() and pmap() vectorise over multiple inputs; at_depth() maps a function at a specified level of nested lists.

• Apply a function conditionally with map_if() (where a predicate returns TRUE) and map_at() (at specific locations).

• Apply a function to slices of a data frame with by_slice(), or to each row with by_row() or map_rows().

• Apply a function to list-elements of a list with lmap(), lmap_if() and lmap_at(). Compared to traditional mapping, the function is applied to x[i] instead of x[[i]], preserving the surrounding list and attributes.

• Reduce a list to a single value by iteratively applying a binary function: reduce() and reduce_right().

• Figure out if a list contains an object: contains().

• Order, sort and split a list based on its components with split_by(), order_by() and sort_by().

List manipulation and creation

• Transpose a list with transpose().

• Create the cartesian product of the elements of several lists with cross_n() and cross_d().

• Flatten a list with flatten().

• Splice lists and other objects with splice().

Predicate functions

(A predicate function is a function that either returns TRUE or FALSE)

• keep() or discard() elements that satisfy the predicate..

• Does every() element or some() elements satisfy the predicate?

• Find the value (detect()) and index (detect_index()) of the first element that satisfies the predicate.

• Find the head/tail that satisfies a predicate: head_while(), tail_while().

Lists of functions

• invoke() every function in a list with given arguments and returns a list, invoke_lgl()/invoke_int()/invoke_dbl()/invoke_chr() returns vectors.

Function operators

• Fill in function arguments with partial().

• Change the way your function takes input with lift() and the lift_xy() family of composition helpers.

• Compose multiple functions into a single function with compose().

• Negate a predicate funtion with negate().

Objects coercion

• Convert an array or matrix to a list with array_tree() and array_branch().

• Convert a list to a vector with as_vector().

Philosophy

The goal is not to try and simulate Haskell in R: purrr does not implement currying or destructuring binds or pattern matching. The goal is to give you similar expressiveness to an FP language, while allowing you to write code that looks and works like R.

• Instead of point free style, use the pipe, %>%, to write code that can be read from left to right.

• Instead of currying, we use ... to pass in extra arguments.

• Anonymous functions are verbose in R, so we provide two convenient shorthands. For unary functions, ~ .x + 1 is equivalent to function(.x) .x + 1. For chains of transformations functions, . %>% f() %>% g() is equivalent to function(.) . %>% f() %>% g().

• R is weakly typed, we need variants map_int(), map_dbl(), etc since we don't know what .f will return.

• R has named arguments, so instead of providing different functions for minor variations (e.g. detect() and detectLast()) I use a named argument, .first. Type-stable functions are easy to reason about so additional arguments will never change the type of the output.

Related work

• rlist, another R package to support working with lists. Similar goals but somewhat different philosophy.

• List operations defined in the Haskell prelude

• Scala's list methods.