Skip to content

Implement axe_fitted.recipe #207

Closed
#208
Closed
Implement axe_fitted.recipe#207
#208

Description

@AshesITR
AshesITR
opened on Nov 25, 2021

It should remove x$template, which contains the prepped data of the training set.

reprex stolen and adapted from tidymodels/recipes#859

library(recipes)
#> Loading required package: dplyr
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
#> 
#> Attaching package: 'recipes'
#> The following object is masked from 'package:stats':
#> 
#>     step
library(butcher)

rec <- recipe(data = dplyr::bind_rows(rep(list(iris), 10)), formula = Species ~ .) %>%
  step_normalize(starts_with("Petal.")) %>%
  step_BoxCox(starts_with("Sepal."))

rec_prepped <- prep(rec)

lobstr::obj_size(rec_prepped)
#> 128,656 B
lobstr::obj_size(butcher(rec_prepped))
#> 70,320 B
lobstr::obj_size(butcher(prep(rec_prepped, retain = FALSE)))
#> 15,936 B

The proposed implementation is quite simple, if I'm not missing anything:

axe_fitted.recipe <- function(x, verbose = FALSE, ...) {
  old <- x
  x$template <- x$template[integer(), ]

  add_butcher_attributes(
    x,
    old,
    verbose = verbose
  )
}

Activity

  1. juliasilge commented on Nov 29, 2021

    @juliasilge
    juliasilge
    on Nov 29, 2021
    Member

    This seems like a really good idea for butchering a recipe, like replacing the template with vctrs::vec_ptype(template):

    library(tidymodels)
#> Registered S3 method overwritten by 'tune':
#>   method                   from   
#>   required_pkgs.model_spec parsnip
data(concrete)

concrete <- 
  concrete %>% 
  group_by(across(-compressive_strength)) %>% 
  summarize(compressive_strength = mean(compressive_strength),
            .groups = "drop")

set.seed(1501)
concrete_split <- initial_split(concrete, strata = compressive_strength)
concrete_train <- training(concrete_split)
concrete_test  <- testing(concrete_split)

rec <- recipe(compressive_strength ~ ., data = concrete_train) %>%
  step_normalize(all_numeric_predictors()) %>% 
  step_poly(all_predictors()) %>% 
  step_interact(~ all_predictors():all_predictors())

prepped <- prep(rec)
bake(prepped, new_data = concrete_test)
#> # A tibble: 249 × 137
#>    compressive_strength cement_poly_1 cement_poly_2 blast_furnace_slag_poly_1
#>                   <dbl>         <dbl>         <dbl>                     <dbl>
#>  1                 4.57       -0.0632        0.0967                    0.0354
#>  2                 7.68       -0.0632        0.0967                    0.0354
#>  3                 7.72       -0.0609        0.0887                    0.0395
#>  4                20.6        -0.0609        0.0887                    0.0395
#>  5                 6.28       -0.0581        0.0794                    0.0440
#>  6                31.0        -0.0581        0.0794                    0.0440
#>  7                10.4        -0.0558        0.0716                    0.0487
#>  8                33.3        -0.0524        0.0611                    0.0584
#>  9                13.7        -0.0521        0.0600                    0.0556
#> 10                 7.51       -0.0511        0.0571                    0.0571
#> # … with 239 more rows, and 133 more variables:
#> #   blast_furnace_slag_poly_2 <dbl>, fly_ash_poly_1 <dbl>,
#> #   fly_ash_poly_2 <dbl>, water_poly_1 <dbl>, water_poly_2 <dbl>,
#> #   superplasticizer_poly_1 <dbl>, superplasticizer_poly_2 <dbl>,
#> #   coarse_aggregate_poly_1 <dbl>, coarse_aggregate_poly_2 <dbl>,
#> #   fine_aggregate_poly_1 <dbl>, fine_aggregate_poly_2 <dbl>, age_poly_1 <dbl>,
#> #   age_poly_2 <dbl>, cement_poly_1_x_cement_poly_2 <dbl>, …

prepped$template <- prepped$template[integer(), ]
juice(prepped)
#> # A tibble: 0 × 137
#> # … with 137 variables: compressive_strength <dbl>, cement_poly_1 <dbl>,
#> #   cement_poly_2 <dbl>, blast_furnace_slag_poly_1 <dbl>,
#> #   blast_furnace_slag_poly_2 <dbl>, fly_ash_poly_1 <dbl>,
#> #   fly_ash_poly_2 <dbl>, water_poly_1 <dbl>, water_poly_2 <dbl>,
#> #   superplasticizer_poly_1 <dbl>, superplasticizer_poly_2 <dbl>,
#> #   coarse_aggregate_poly_1 <dbl>, coarse_aggregate_poly_2 <dbl>,
#> #   fine_aggregate_poly_1 <dbl>, fine_aggregate_poly_2 <dbl>, …
bake(prepped, new_data = concrete_test)
#> # A tibble: 249 × 137
#>    compressive_strength cement_poly_1 cement_poly_2 blast_furnace_slag_poly_1
#>                   <dbl>         <dbl>         <dbl>                     <dbl>
#>  1                 4.57       -0.0632        0.0967                    0.0354
#>  2                 7.68       -0.0632        0.0967                    0.0354
#>  3                 7.72       -0.0609        0.0887                    0.0395
#>  4                20.6        -0.0609        0.0887                    0.0395
#>  5                 6.28       -0.0581        0.0794                    0.0440
#>  6                31.0        -0.0581        0.0794                    0.0440
#>  7                10.4        -0.0558        0.0716                    0.0487
#>  8                33.3        -0.0524        0.0611                    0.0584
#>  9                13.7        -0.0521        0.0600                    0.0556
#> 10                 7.51       -0.0511        0.0571                    0.0571
#> # … with 239 more rows, and 133 more variables:
#> #   blast_furnace_slag_poly_2 <dbl>, fly_ash_poly_1 <dbl>,
#> #   fly_ash_poly_2 <dbl>, water_poly_1 <dbl>, water_poly_2 <dbl>,
#> #   superplasticizer_poly_1 <dbl>, superplasticizer_poly_2 <dbl>,
#> #   coarse_aggregate_poly_1 <dbl>, coarse_aggregate_poly_2 <dbl>,
#> #   fine_aggregate_poly_1 <dbl>, fine_aggregate_poly_2 <dbl>, age_poly_1 <dbl>,
#> #   age_poly_2 <dbl>, cement_poly_1_x_cement_poly_2 <dbl>, …

    Created on 2021-11-29 by the reprex package (v2.0.1)

  2. juliasilge commented on Nov 30, 2021

    @juliasilge
    juliasilge
    on Nov 30, 2021
    Member

    @AshesITR would you be interested in contributing a PR to hardhat to implement this butcher method for a recipe? We have an article here with some advice on contributing to butcher, but like you have probably already discovered, the method would go in this file.

  3. AshesITR commented on Nov 30, 2021

    @AshesITR
    AshesITR
    on Nov 30, 2021
    ContributorAuthor

    Sure, I'll make this a PR. Regarding df[integer(), ] vs. vctrs::vec_ptype(df): Do you have an opinion regarding any of these alternatives?

  4. juliasilge commented on Nov 30, 2021

    @juliasilge
    juliasilge
    on Nov 30, 2021
    Member

    Do you have an opinion on that @DavisVaughan? butcher doesn't currently import vctrs but does import tibble, which imports vctrs.

  5. AshesITR
    added a commit that references this issue on Nov 30, 2021

    Implement axe_fitted.recipe

    f2e2534
  6. juliasilge
    closed this as completedin #208on Dec 2, 2021
  7. juliasilge
    added a commit that references this issue on Dec 2, 2021

    Implement axe_fitted.recipe (#208)

    1a49b9b

  8. github-actions commented on Dec 17, 2021

    @github-actions
    github-actionsbot
    on Dec 17, 2021
    Contributor

    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.

  9. github-actions
    locked and limited conversation to collaborators on Dec 17, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions