Challenges using `all_nominal()` and `all_numeric()`

[juliasilge]

We have lots of examples of folks having trouble with all_nominal() and all_numeric() when they accidentally select outcomes or other non-predictors. Mine talked about running into this during teaching, and here is one example of many from the general user base. A significant challenge is that because of how/when these steps are evaluated, the error message is often opaque.

What would be a good way to reduce this friction?

• Should all_nominal() and all_numeric() only select predictors?
• Should these be replaced by new selectors like all_nominal_preds() and all_numeric_preds()?
• Some recipe steps are more prone to this problem, like step_dummy(). Protections at the step level?
• Something else?

[mine-cetinkaya-rundel]

I like the idea of protections at the step level, but if this is not feasible/desirable (or even if it is), I think adding selectors like all_nominal_preds() and all_numeric_preds() and promoting these in the function docs can help. I think it's also useful to demonstrate using multiple selectors in the function docs (e.g. step_dummy(all_nominal(), all_predictors(), -all_outcomes())) though the need for that is lower if the combined selectors are in place.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org