address some julia 1.8 type param warnings; partially address #835

src/measures/finite.jl

@@ -530,12 +530,7 @@ for M in (:MulticlassTruePositiveRate, :MulticlassTrueNegativeRate,
             average::T
             return_type::Type{U}
         end
-        # @ablaom says next line looks redundant:
-        $M(average::T, return_type::Type{U}) where {T, U} =
-            $M(average, return_type)
-        $M(; average=macro_avg,
-           return_type=LittleDict) where T<:MulticlassAvg where
-           U<:Union{Vector, LittleDict} = $M(average, return_type)
+        $M(; average=macro_avg, return_type=LittleDict) = $M(average, return_type)
     end
     eval(ex)
 end

src/measures/probabilistic.jl

@@ -88,13 +88,12 @@ end
 # Missing values not supported, but allow `Missing` in eltype, because
 # `skipinvalid(yhat, y)` does not tighten the type. See doc string above.
 
-call(::AUC, ŷ::ArrMissing{UnivariateFinite{S,V,R,P}}, y) where {S,V,R,P} =
+call(::AUC, ŷ::ArrMissing{UnivariateFinite{S,V,R,P}}, y) where {S,V,R,P<:Real} =
     _auc(P, ŷ, y)
 
-# corner case of UnivariateFinite's of mixed type
-call(::AUC, ŷ::ArrMissing{UnivariateFinite}, y) where {S,V,R,P} =
-    _auc(Float64, ŷ, y)
-
+# corner case of UnivariateFinite's of mixed type; next line could be removed after
+# resolution of https://github.com/JuliaAI/CategoricalDistributions.jl/issues/37
+call(::AUC, ŷ::ArrMissing{UnivariateFinite}, y) =  _auc(Float64, ŷ, y)
 
 
 # ========================================================