Verify model assumptions with tidymodels

Question

Outside of the tidymodels universe, it's easy to verify model assumptions. For example with linear regression (function lm), the package performance create understandable graphics and easy functions (check_heteroscedasticity()) to verify assumptions of a linear regression model :

normality of residuals
independence of the residuals
homogeneity of the residuals
non-collinearity of the variables.

Is there in the tidymodels universe equivalent packages to verify assumptions of a model ? Tidymodels packages create parnsnip object, so old model evaluation packages like performance are useless.

Thank you for your help

I would say that the [`infer`](https://infer.netlify.app) is designed for that purpose; have a look at the [`hypothesise`](https://infer.netlify.app/reference/hypothesize.html) function. — Konrad, Sep 06 '21 at 12:58
Note that your terminology is not „correct“, i.e. you don‘t want to check your hypotheses, you want to check your model assumptions (!). — deschen, Sep 06 '21 at 13:38

score 7 · Accepted Answer · answered Sep 07 '21 at 02:08

The fitted models you get from using {parsnip} or other {tidymodels} will contain the underlying fitted model for whatever engine you are using.

Sometimes the fitted parsnip object will work directly with the function you are using. This is the case with the check_model() function from {performance}.

library(tidymodels)
library(performance)

lm_spec <- linear_reg() %>%
  set_mode("regression") %>%
  set_engine("lm")

lm_fit <- fit(lm_spec, mpg ~ ., data = mtcars)

lm_fit %>%
  check_model()

Other times you will get an error because the function doesn't know what to do with a model_fit object. You can use the extract_fit_engine() which will extract the fit produced by the engine, that can then be used with check_heteroscedasticity().

lm_fit %>% 
  extract_fit_engine() %>%
  check_heteroscedasticity()
#> OK: Error variance appears to be homoscedastic (p = 0.188).

^{Created on 2021-09-06 by the reprex package (v2.0.1)}

Verify model assumptions with tidymodels

1 Answers1