Enforce empty dots?

[hfrick]

Currently, yardstick has a lot of dots documented as "Currently not used". Have we already looked into enforcing empty dots in such cases and decided against that? Otherwise, this would be a nice improvement.

library(yardstick)
data("two_class_example")

j_index(two_class_example, truth, predicted, hello = "world")
#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 j_index binary         0.673

^{Created on 2025-12-09 with reprex v2.1.1}

[EmilHvitfeldt]

We can't do it fully but we can do some of it. The main issue comes from the way metric_set() works.

Imagine the following code

library(yardstick)

set <- metric_set(accuracy, recall, roc_auc)

set(two_class_example, truth, Class1, estimate = predicted)

when we are inside the metric_set() infrastructure we create some calls that we then evaluate. These calls looks something like this

$accuracy
(function(data, ...) {
  UseMethod("accuracy")
})(data = ~data, truth = ~truth, estimate = ~predicted, estimator = ~estimator, 
    na_rm = ~na_rm, event_level = ~event_level, case_weights = ~NULL)

$recall
(function(data, ...) {
  UseMethod("recall")
})(data = ~data, truth = ~truth, estimate = ~predicted, estimator = ~estimator, 
    na_rm = ~na_rm, event_level = ~event_level, case_weights = ~NULL)

and it is at this point where it is kinda hard to remove the arguments that we don't need, as the function accuracy has then signature function(data, ...). Could we do some stuff pull the accuracy.data.frame method.

we could for example turn the following line of code

yardstick/R/aaa-metrics.R

Line 424 in 4cb3890

class_calls <- lapply(class_fns, call2, !!!class_args)

into

      class_calls <- list()
      for (i in seq_along(class_fns)) {
        args <- class_args
        needed_args <- names(formals(getS3method(
          names(class_fns)[[i]],
          "data.frame"
        )))
        needed_args <- setdiff(needed_args, "...")
        class_calls[[i]] <- call2(class_fns[[i]], !!!args[needed_args])
      }
      names(class_calls) <- names(class_fns)

but even then there are some edge cases it doesn't cover.

We can however check in the _vec() functions with no issue

[EmilHvitfeldt]

Doing a little bit more work and the way metric_tweak() is implemented actually ends up messing up with other methods as well. So I'm going to close this issue as it isn't the biggest deal to have this, more of a "nice to have"