workflows confuses an 'id' special for dplyr::id()

[topepo]

See https://stackoverflow.com/questions/79163528/multistate-survival-models-with-censored-and-tidymodels

This might end up being a parsnip issue but we should start here.

library(tidyverse)
library(tidymodels)
library(censored)
#> Loading required package: survival

df <- tibble(
  sample_id = seq(1:1000),
  sex = sample(c("Male","Female"), 1000, replace = T),
  duration = rnorm(1000, mean = 75, sd = 7),
  outcome = sample(c("censor","Lost","Won"), 1000, replace = T)
  
) %>%
  mutate(
    outcome = factor(outcome)
  )

mod_df <- df %>%
  mutate(
    sale_surv = Surv(duration, outcome),
    .keep = "unused"
  )

set.seed(1988)

mod_split <- initial_split(mod_df)
mod_train <- training(mod_split)
mod_test <- testing(mod_split)

mod_rec <- recipe(sale_surv ~ sex, data = mod_train) # no `id` here 

mod_spec <- proportional_hazards() %>%
  set_mode("censored regression") %>%
  set_engine("survival") 

mod_wflow <- workflow() %>%
  add_recipe(mod_rec) %>%
  add_model(mod_spec, formula = sale_surv ~ sex + id(sample_id)) # add a formula here

mod_fit <- fit(mod_wflow, data = mod_train)
#> Error:
#> ! `id()` was deprecated in dplyr 0.5.0 and is now defunct.
#> ℹ Please use `vctrs::vec_group_id()` instead.

[topepo]

Tracing the error, it happens when .fit_model() invokes fit.action_model(). In there, the formula is correctly extracted:

sale_surv ~ sex + id(sample_id)
<environment: 0x12ee76170>

From here, it calls fit_from_formula() when then calls fit.model_spec() via

fit(spec, formula = formula, data = data, case_weights = case_weights, 
    control = control_parsnip)

This leads to calling

survival::coxph(formula = sale_surv ~ sex + id(sample_id), data = data, 
                x = TRUE, model = TRUE)

id is a “formula special” and isn’t a function in the survival package (so using the conflicted isn’t a solution).

Note that the same call to results in the same error when run in the global environment:

> survival::coxph(formula = sale_surv ~ sex + id(sample_id), data = mod_df, 
+                 x = TRUE, model = TRUE)
Error:
! `id()` was deprecated in dplyr 0.5.0 and is now defunct.
ℹ Please use `vctrs::vec_group_id()` instead.
Run `rlang::last_trace()` to see where the error occurred.