The kulczynski1 function is documented to return a dissimilarity score, but its formula actually returns a similarity score.
I expect such contradiction to lead to awkward results when used, but there appears to be no report of related issue in the past three years since the function is added. This implies that either users are happy with a similarity score being returned, or that the function is not really being used. I’m inclined to believe the latter is the reason.
Trying to “fix” the metric would require “flipping” the result, which would be a drastic change. Since probably nobody actually uses the function, I suggest to deprecate the function and remove it later. I opened gh-21187 to collect feedback.
Do you think it’s appropriate to deprecate and later remove the kulczynski1 function?