vec_split()

[hadley]

vec_split <- function(x, f) {
  unname(split(x, vec_duplicate_id(f)))
}

But implemented more efficiently internally, and using vec_slice()

[hadley]

Or would it return a tibble? One row would be the keys (vec_unique()) and the other would be the values (as above). Is this the answer to #197?

[DavisVaughan]

This can do some pretty neat things, especially if using data frames as f

suppressPackageStartupMessages({
  library(vctrs)
  library(gapminder)
  library(tidyr)
  library(dplyr)
})
#> Warning: package 'dplyr' was built under R version 3.5.2

vec_split <- function(x, f) {
  keys <- vec_unique(f)
  
  prxy <- vec_duplicate_id(f)
  prxy_keys <- vec_unique(prxy)
  
  values <- vctrs:::map(prxy_keys, function(prxy_key) {
    vec_slice(x, vec_equal(prxy, prxy_key))
  })
  
  tibble::tibble(.keys = keys, .values = values)
}

vec_split(iris, iris$Species)
#> # A tibble: 3 x 2
#>   .keys      .values              
#>   <fct>      <list>               
#> 1 setosa     <data.frame [50 × 5]>
#> 2 versicolor <data.frame [50 × 5]>
#> 3 virginica  <data.frame [50 × 5]>

vec_split(iris, 1)
#> # A tibble: 1 x 2
#>   .keys .values               
#>   <dbl> <list>                
#> 1     1 <data.frame [150 × 5]>

vec_split(iris, NA)
#> # A tibble: 1 x 2
#>   .keys .values               
#>   <lgl> <list>                
#> 1 NA    <data.frame [150 × 5]>

vec_split(iris, NULL)
#> Error: All columns in a tibble must be 1d or 2d objects:
#> * Column `.keys` is NULL
#> Backtrace:
#>     █
#>  1. └─global::vec_split(iris, NULL)
#>  2.   └─tibble::tibble(.keys = keys, .values = values)
#>  3.     └─tibble:::lst_to_tibble(xlq$output, .rows, .name_repair, lengths = xlq$lengths)
#>  4.       └─tibble:::check_valid_cols(x)

# split by unique combinations of 2 columns
gap_nest <- nest(gapminder, continent, country)
vec_split(gap_nest, gap_nest$data)
#> # A tibble: 142 x 2
#>    .keys            .values          
#>    <list>           <list>           
#>  1 <tibble [1 × 2]> <tibble [12 × 5]>
#>  2 <tibble [1 × 2]> <tibble [12 × 5]>
#>  3 <tibble [1 × 2]> <tibble [12 × 5]>
#>  4 <tibble [1 × 2]> <tibble [12 × 5]>
#>  5 <tibble [1 × 2]> <tibble [12 × 5]>
#>  6 <tibble [1 × 2]> <tibble [12 × 5]>
#>  7 <tibble [1 × 2]> <tibble [12 × 5]>
#>  8 <tibble [1 × 2]> <tibble [12 × 5]>
#>  9 <tibble [1 × 2]> <tibble [12 × 5]>
#> 10 <tibble [1 × 2]> <tibble [12 × 5]>
#> # … with 132 more rows

# or this way
vec_split(gapminder, select(gapminder, continent, country))
#> # A tibble: 142 x 2
#>    .keys$continent $country    .values          
#>    <fct>           <fct>       <list>           
#>  1 Asia            Afghanistan <tibble [12 × 6]>
#>  2 Europe          Albania     <tibble [12 × 6]>
#>  3 Africa          Algeria     <tibble [12 × 6]>
#>  4 Africa          Angola      <tibble [12 × 6]>
#>  5 Americas        Argentina   <tibble [12 × 6]>
#>  6 Oceania         Australia   <tibble [12 × 6]>
#>  7 Europe          Austria     <tibble [12 × 6]>
#>  8 Asia            Bahrain     <tibble [12 × 6]>
#>  9 Asia            Bangladesh  <tibble [12 × 6]>
#> 10 Europe          Belgium     <tibble [12 × 6]>
#> # … with 132 more rows

mat <- matrix(1:50, 10, 5)
mat_f <- rep(1:2, times = 5)
vec_split(mat, mat_f)
#> # A tibble: 2 x 2
#>   .keys .values      
#>   <int> <list>       
#> 1     1 <int [5 × 5]>
#> 2     2 <int [5 × 5]>

^{Created on 2019-02-27 by the reprex package (v0.2.1.9000)}

[hadley]

And the fact that the implementation is so simple is a good signal that the underlying primitives are correct. At some point, we will need to rewrite in C to avoid creating the internal dictionary twice (once for the unique values and once for the duplicates).