parsnip should raise an error when users use logistic_reg() for multiclass classification

[zenggyu]

Currently parsnip does not throw an error if a user chooses logistic_reg() to fit a multiclass classification model, but the result is wrong and misleading. Moreover, the documentation doesn't mention that logistic_reg() should only be used for binary classification. It took me a while to figure out that I should use multinom_reg() instead. It would be nice if parsnip can prevent this usage explicitly.

library(parsnip)
fit <- logistic_reg() %>%
  set_mode("classification") %>%
  set_engine("glm") %>%
  fit(Species ~ ., data = iris)
#> Warning: glm.fit: algorithm did not converge
#> Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
predict(fit, new_data = iris)
#> # A tibble: 150 x 1
#>    .pred_class
#>    <fct>      
#>  1 setosa     
#>  2 setosa     
#>  3 setosa     
#>  4 setosa     
#>  5 setosa     
#>  6 setosa     
#>  7 setosa     
#>  8 setosa     
#>  9 setosa     
#> 10 setosa     
#> # … with 140 more rows

^{Created on 2021-08-23 by the reprex package (v2.0.0)}

[juliasilge]

We recently revamped our documentation so that logistic_reg() says:

logistic_reg() defines a generalized linear model for binary outcomes. A linear combination of the predictors is used to model the log odds of an event.

We are already highlighting that it is for binary outcomes but we could add a pointer to multinom_reg() on that landing page.

Sidenote: Should we also mention that some engines like xgboost and ranger handle multiclass classification on their own? Probably not here.

In translate.logistic_reg we could check for a two-factor outcome.

[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.