Applying functions designed to take base R model objects on Parsnip objects?

Question

Most package authors design functions to take in the traditional R model objects like is there anyway to feed them a model object created by parsnip package and tidymodels?

an example:

model <- lm(mpg ~ am + cyl + wt + hp, data = mtcars)
car::avPlots(model, terms = . ~ am + cyl + wt + hp)

but if you create the model through tidymodels it won't work since the output object is different#fit_ex5_1 <-

model_pars <-
  linear_reg() %>% 
  set_engine("lm") %>% 
  fit(mpg ~ am + cyl + wt + hp, data = mtcar)

#this doesn't work:
car::avPlots(model_pars, terms = . ~ am + cyl + wt + hp)

Answer 1

I imagine there may be various methods depending on the type of model. In the case of lm() you can extract parsnip_model$fit, which is an object of class lm, like the output of lm():

parsnip_model <- linear_reg() |>
    fit(mpg ~ am + cyl + wt + hp, data = mtcars)

class(parsnip_model$fit) # lm

# Compare the models excl. [["call"]] which will be different
all.equal(
    parsnip_model$fit[names(parsnip_model$fit) != "call"],
    model[names(model) != "call"]
) # TRUE

car::avPlots(parsnip_model$fit) # this works

Note: I did not include the terms argument to the car::avPlots() call as the plot includes all the terms by default, but the code runs if you include them.

Answer 2

There is a function, extract_fit_engine(), that will return the underlying model object if you have a parsnip model fit or a fitted workflow.

library(tidymodels)

linear_reg() %>% 
  set_engine("lm") %>% 
  fit(mpg ~ am + cyl + wt + hp, data = mtcars) %>% 
  extract_fit_engine()
#> 
#> Call:
#> stats::lm(formula = mpg ~ am + cyl + wt + hp, data = data)
#> 
#> Coefficients:
#> (Intercept)           am          cyl           wt           hp  
#>    36.14654      1.47805     -0.74516     -2.60648     -0.02495

^{Created on 2023-09-06 with reprex v2.0.2}