augmenting regression models should add the residual if the outcome is available

[topepo]

Other augment() methods will add a residual column.

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

set.seed(4325)
data_split <- initial_split(mpg)
data_train <- training(data_split)
data_test  <- testing(data_split)

milage_recipe <- 
  recipe(hwy ~ manufacturer + year, data = data_train) %>%
  step_other(manufacturer, threshold = 0.1) %>%
  step_dummy(manufacturer) %>%
  step_interact(~ year:starts_with("manufacturer"))

wflow <- 
  workflow() %>%
  add_model(linear_reg() %>% set_engine("lm")) %>%
  add_recipe(milage_recipe)

wflow %>%
  fit(data_train) %>%
  augment(data_train)
#> # A tibble: 175 × 12
#>    manufacturer model    displ  year   cyl trans  drv     cty   hwy fl    class 
#>    <chr>        <chr>    <dbl> <int> <int> <chr>  <chr> <int> <int> <chr> <chr> 
#>  1 dodge        ram 150…   4.7  2008     8 auto(… 4        13    17 r     pickup
#>  2 toyota       corolla    1.8  2008     4 auto(… f        26    35 r     compa…
#>  3 nissan       altima     2.4  1999     4 manua… f        21    29 r     compa…
#>  4 volkswagen   passat     2    2008     4 manua… f        21    29 p     midsi…
#>  5 toyota       4runner…   3.4  1999     6 auto(… 4        15    19 r     suv   
#>  6 ford         mustang    4.6  1999     8 auto(… r        15    21 r     subco…
#>  7 volkswagen   passat     2    2008     4 auto(… f        19    28 p     midsi…
#>  8 toyota       toyota …   2.7  2008     4 manua… 4        17    22 r     pickup
#>  9 volkswagen   passat     2.8  1999     6 auto(… f        16    26 p     midsi…
#> 10 volkswagen   passat     1.8  1999     4 manua… f        21    29 p     midsi…
#> # … with 165 more rows, and 1 more variable: .pred <dbl>

^{Created on 2021-12-18 by the reprex package (v2.0.1)}

[DavisVaughan]

Ideally we'd let parsnip's augment() method handle adding on the residuals, which it does try and do if the outcome column is there. But it relies on this fit$preproc$y_var information which doesn't seem to be there for some reason when the model is fit through a workflow. So even if we passed the outcome data through to the parsnip augment() method, it doesn't seem like it would add the residuals column on because that info isn't there.

library(workflows)
library(parsnip)
data("mpg", package = "ggplot2")

wf <- workflow() %>%
  add_variables(hwy, c(displ, year)) %>%
  add_model(linear_reg())

wf <- wf %>%
  fit(mpg)

fit_parsnip <- extract_fit_parsnip(wf)

# Where is `y_var`?
fit_parsnip$preproc
#> $x_var
#> [1] "displ" "year" 
#> 
#> $y_var
#> character(0)


parsnip:::augment.model_fit
#> function (x, new_data, ...) 
#> {
#>     ret <- new_data
#>     if (x$spec$mode == "regression") {
#>         check_spec_pred_type(x, "numeric")
#>         ret <- ret %>% dplyr::bind_cols(predict(x, new_data = new_data))
#>         if (length(x$preproc$y_var) > 0) {
#>             y_nm <- x$preproc$y_var
#>             if (any(names(new_data) == y_nm)) {
#>                 ret <- dplyr::mutate(ret, .resid = !!rlang::sym(y_nm) - 
#>                   .pred)
#>             }
#>         }
#>     }
#>     else if (x$spec$mode == "classification") {
#>         if (spec_has_pred_type(x, "class")) {
#>             ret <- dplyr::bind_cols(ret, predict(x, new_data = new_data, 
#>                 type = "class"))
#>         }
#>         if (spec_has_pred_type(x, "prob")) {
#>             ret <- dplyr::bind_cols(ret, predict(x, new_data = new_data, 
#>                 type = "prob"))
#>         }
#>     }
#>     else {
#>         rlang::abort(paste("Unknown mode:", x$spec$mode))
#>     }
#>     as_tibble(ret)
#> }
#> <bytecode: 0x7fbdd3d9fb40>
#> <environment: namespace:parsnip>

^{Created on 2022-02-24 by the reprex package (v2.0.1)}

[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.