You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As noted here, the development version of {yardstick} includes a function, ranked_prob_score(), for use with ordinal models. {ordered} should be harmonized with its use, however it plays out.
What makes the most sense to me now is to (1) write a standalone script that illustrates the use of several ordinal-appropriate metrics already in {yardstick} and, once it's stress-tested, (2) decide whether to fashion it into an example set, vignette, article or other format. If yardstick limitations arise, then (3) come back to #7 (and, if not, then close that issue). The metrics should include some "numeric"s (for those who want to use them), some that rely only on class predictions (for engines that don't support probability predictions), and some "tweaked" metrics (e.g. kap_quad), as well as those that rely on class probabilities.
Activity
corybrunson commentedon Oct 5, 2026
What makes the most sense to me now is to (1) write a standalone script that illustrates the use of several ordinal-appropriate metrics already in {yardstick} and, once it's stress-tested, (2) decide whether to fashion it into an example set, vignette, article or other format. If yardstick limitations arise, then (3) come back to #7 (and, if not, then close that issue). The metrics should include some "numeric"s (for those who want to use them), some that rely only on class predictions (for engines that don't support probability predictions), and some "tweaked" metrics (e.g.
kap_quad), as well as those that rely on class probabilities.