Behaviour of features with functions that return argument > length 1

[njtierney]

I've noticed that when passing features a function that outputs something of lenght > 1, you get a long warning of renaming - I just wanted to check if this was expected behavhouiur? It's quite cool that this just works! I can get as many columns as there are outputs from the function. But just wanted to raise it as I hasn't seen it documented.

library(brolgar)
# Passing a vector to `mean` vs `diff` 
# mean
mean(c(1:10))
#> [1] 5.5
# input = vector of length 10
# output = vector of lenght 1

diff(c(1:10))
#> [1] 1 1 1 1 1 1 1 1 1
# input = vector of length 10
# input = vector of length 9

# so consequently, passing `features` a function that produces a vector
# will give you quite different return formats.

# mean
wages %>%
  features(ln_wages, 
           list(mean = mean))
#> # A tibble: 888 x 2
#>       id  mean
#>    <int> <dbl>
#>  1    31  1.75
#>  2    36  2.33
#>  3    53  1.89
#>  4   122  2.17
#>  5   134  2.48
#>  6   145  1.76
#>  7   155  2.17
#>  8   173  1.93
#>  9   206  2.27
#> 10   207  2.11
#> # … with 878 more rows

# range
wages %>%
  features(ln_wages, 
           list(range = range))
#> New names:
#> * `.?` -> `.?...1`
#> * `.?` -> `.?...2`
#> New names:
#> * `diff_.?...12` -> `diff_.?...13`
#> New names:
#> * `diff_.?...1` -> `diff_.?...2`
#> * `diff_.?...2` -> `diff_.?...3`
#> * `diff_.?...3` -> `diff_.?...4`
#> * `diff_.?...4` -> `diff_.?...5`
#> * `diff_.?...5` -> `diff_.?...6`
#> * ...
#> # A tibble: 888 x 14
#>       id `diff_.?...2` `diff_.?...3` `diff_.?...4` `diff_.?...5` `diff_.?...6`
#>    <int>         <dbl>         <dbl>         <dbl>         <dbl>         <dbl>
#>  1    31        -0.058         0.036         0.28          0.182        -0.222
#>  2    36        -0.184         0.458         0.317        -0.754         1.11 
#>  3    53        -0.225         1.70         -1.65         -0.03          0.316
#>  4   122         0.806        -1.00         -1.16          1.68         -0.263
#>  5   134         0.095         0.224         0.064        -0.124         0.063
#>  6   145        -0.086         0.12          0.289         0.151        -0.265
#>  7   155         0.302        -0.543         0.627         0.158        -0.593
#>  8   173         0.134         0.283         0.051        -0.011         0.319
#>  9   206         0.269         0.185        NA            NA            NA    
#> 10   207         0.169         0.182        -0.068         0.361        -0.172
#> # … with 878 more rows, and 8 more variables: `diff_.?...7` <dbl>,
#> #   `diff_.?...8` <dbl>, `diff_.?...9` <dbl>, `diff_.?...10` <dbl>,
#> #   `diff_.?...11` <dbl>, `diff_.?...12` <dbl>, diff <dbl>,
#> #   `diff_.?...14` <dbl>

^{Created on 2020-08-19 by the reprex package (v0.3.0)}

[mitchelloharawild]

All values of the vector needs names if multiple values will be returned. There shouldn't be this many warnings (one per column is appropriate), but this is how I would use range():

library(brolgar)
# range
wages %>%
  features(ln_wages, 
           list(range = ~ setNames(range(.), c("min", "max"))))
#> # A tibble: 888 x 3
#>       id range_min range_max
#>    <int>     <dbl>     <dbl>
#>  1    31     1.43       2.13
#>  2    36     1.80       2.93
#>  3    53     1.54       3.24
#>  4   122     0.763      2.92
#>  5   134     2.00       2.93
#>  6   145     1.48       2.04
#>  7   155     1.54       2.64
#>  8   173     1.56       2.34
#>  9   206     2.03       2.48
#> 10   207     1.58       2.66
#> # … with 878 more rows

^{Created on 2020-08-20 by the reprex package (v0.3.0)}

[njtierney]

Neat! Thanks for that! So there isn't currently a way to specify names within features?

Also if it is helpful I'd be happy to document this behaviour in fabletools, since I'll be doing it for brolgar as well

[mitchelloharawild]

More documentation is always welcomed.

How do you mean by 'specify names within features'? Do you have a proposed interface improvement?