error informatively with bad object in set_* functions

[simonpcouch]

Closes #776.

Worth noting that this error message has changed since the issue was filed, following up on #769.

library(parsnip)

bag_tree %>% set_mode("classification")
#> Error in UseMethod("set_mode"): no applicable method for 'set_mode' applied to an object of class "function"

^{Created on 2022-08-15 by the reprex package (v2.0.1)}

The errors now look like:

library(parsnip)

bag_tree %>% set_mode("classification")
#> Error in `set_mode()`:
#> ! `set_mode()` expected a model specification to be supplied to the
#>   `object` argument, but received a(n) `function` object.
#> ℹ Did you mistakenly pass `model_function` rather than `model_function()`?

bag_tree %>% set_engine("rpart")
#> Error in `set_engine()`:
#> ! `set_engine()` expected a model specification to be supplied to the
#>   `object` argument, but received a(n) `function` object.
#> ℹ Did you mistakenly pass `model_function` rather than `model_function()`?

# won't raise "info" part of error if not a parsnip-namespaced function
# not a function
1L %>% set_args(mode = "classification")
#> Error in `set_args()`:
#> ! `set_args()` expected a model specification to be supplied to the
#>   `object` argument, but received a(n) `integer` object.

# not from parsnip
cos %>% set_mode("classification")
#> Error in `set_mode()`:
#> ! `set_mode()` expected a model specification to be supplied to the
#>   `object` argument, but received a(n) `function` object.

^{Created on 2022-08-15 by the reprex package (v2.0.1)}

Note that, in the case where a user supplies a parsnip-namespaced function that's not a model type function, the error message is misleading. If we wanted to make the logic for that part of the error more exact, we could class the model type functions with a subclass!

[simonpcouch]

Requesting your review here, @EmilHvitfeldt, since this messaging is relevant to tidyclust, but no worries on reviewing this before the end of your vacation! These changes won't be merged until after the upcoming release of parsnip anyway, so reviewing in September is great. :)

I use "model specification" in the prompt rather than a specific object class so that this will hopefully generalize well to tidyclust. Feel free to suggest improvements if you think there's a more helpful way of phrasing this, though.

[EmilHvitfeldt]

Overall good implementation. 1 question in review

[simonpcouch]

Looping in @topepo before merging here. See Emil's comment above—does it make more sense to just drop the second bit of the prompt entirely? Or should we think about subclassing the model spec functions so that this logic is solid?

[topepo]

I agree to leave it as-is (until we see it as a big issue)

[github-actions]

This pull request 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.