Have steps return integers when appropriate

[EmilHvitfeldt]

Some steps that return what are essentially integers are returning them as doubles. It might be helpful to handle the conversion to help with object sizes.

library(recipes)
library(modeldata)

data("Chicago")

recipe(ridership ~ date, data = Chicago) %>%
  step_date(date, label = FALSE, keep_original_cols = FALSE) %>%
  prep() %>%
  bake(new_data = NULL)
#> # A tibble: 5,698 × 4
#>    ridership date_dow date_month date_year
#>        <dbl>    <dbl>      <dbl>     <dbl>
#>  1     15.7         2          1      2001
#>  2     15.8         3          1      2001
#>  3     15.9         4          1      2001
#>  4     15.9         5          1      2001
#>  5     15.4         6          1      2001
#>  6      2.42        7          1      2001
#>  7      1.47        1          1      2001
#>  8     15.5         2          1      2001
#>  9     15.9         3          1      2001
#> 10     15.9         4          1      2001
#> # … with 5,688 more rows

recipe(ridership ~ date, data = Chicago) %>%
  step_date(date, keep_original_cols = FALSE) %>%
  step_dummy(all_nominal_predictors()) %>%
  prep() %>%
  bake(new_data = NULL)
#> # A tibble: 5,698 × 19
#>    ridership date_year date_dow_Mon date_dow_Tue date_dow_Wed date_dow_Thu
#>        <dbl>     <dbl>        <dbl>        <dbl>        <dbl>        <dbl>
#>  1     15.7       2001            1            0            0            0
#>  2     15.8       2001            0            1            0            0
#>  3     15.9       2001            0            0            1            0
#>  4     15.9       2001            0            0            0            1
#>  5     15.4       2001            0            0            0            0
#>  6      2.42      2001            0            0            0            0
#>  7      1.47      2001            0            0            0            0
#>  8     15.5       2001            1            0            0            0
#>  9     15.9       2001            0            1            0            0
#> 10     15.9       2001            0            0            1            0
#> # … with 5,688 more rows, and 13 more variables: date_dow_Fri <dbl>,
#> #   date_dow_Sat <dbl>, date_month_Feb <dbl>, date_month_Mar <dbl>,
#> #   date_month_Apr <dbl>, date_month_May <dbl>, date_month_Jun <dbl>,
#> #   date_month_Jul <dbl>, date_month_Aug <dbl>, date_month_Sep <dbl>,
#> #   date_month_Oct <dbl>, date_month_Nov <dbl>, date_month_Dec <dbl>

^{Created on 2021-08-10 by the reprex package (v2.0.1)}

[juliasilge]

FWIW the current behavior mirrors what happens in model.matrix() (in fact, that is why they turn out double):

library(tidyverse)
data(penguins, package = "palmerpenguins")
model.matrix(bill_length_mm ~ species + sex + bill_depth_mm, 
             data = penguins) %>%
  as_tibble()
#> # A tibble: 333 × 5
#>    `(Intercept)` speciesChinstrap speciesGentoo sexmale bill_depth_mm
#>            <dbl>            <dbl>         <dbl>   <dbl>         <dbl>
#>  1             1                0             0       1          18.7
#>  2             1                0             0       0          17.4
#>  3             1                0             0       0          18  
#>  4             1                0             0       0          19.3
#>  5             1                0             0       1          20.6
#>  6             1                0             0       0          17.8
#>  7             1                0             0       1          19.6
#>  8             1                0             0       0          17.6
#>  9             1                0             0       1          21.2
#> 10             1                0             0       1          21.1
#> # … with 323 more rows

^{Created on 2021-08-16 by the reprex package (v2.0.1)}

Thinking about how often a model that needs dummy variables really can use integers vs. real numbers might be worthwhile.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex https://reprex.tidyverse.org) and link to this issue.