diff --git a/Project.toml b/Project.toml
index 2b7c7729..438c1337 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "MLJBase"
 uuid = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
 authors = ["Anthony D. Blaom <anthony.blaom@gmail.com>"]
-version = "0.15.1"
+version = "0.15.2"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
diff --git a/src/MLJBase.jl b/src/MLJBase.jl
index 4937a6fe..cda90d22 100644
--- a/src/MLJBase.jl
+++ b/src/MLJBase.jl
@@ -186,7 +186,7 @@ export orientation, reports_each_observation,
     spports_weights, prediction_type
 
 # measures/continuous.jl:
-export mav, mae, mape, rms, rmsl, rmslp1, rmsp, l1, l2
+export mav, mae, mape, rms, rmsl, rmslp1, rmsp, l1, l2, log_cosh
 
 # measures/confusion_matrix.jl:
 export confusion_matrix, confmat
diff --git a/src/measures/continuous.jl b/src/measures/continuous.jl
index b40294e7..39045443 100644
--- a/src/measures/continuous.jl
+++ b/src/measures/continuous.jl
@@ -320,3 +320,33 @@ function (m::MAPE)(ŷ::Vec{<:Real}, y::Vec{<:Real})
     end
     return ret / count
 end
+
+struct LogCosh <: Measure end
+
+"""
+    log_cosh(ŷ, y)
+
+Log-Cosh per-observation loss:
+
+``\\text{Log-Cosh Loss} = log(cosh(ŷᵢ-yᵢ))``
+
+where `yᵢ` is the target and `ŷᵢ` is the output.
+
+For more information, run `info(log_cosh)`.
+"""
+const log_cosh = LogCosh()
+
+metadata_measure(LogCosh;
+    name                     = "log_cosh",
+    target_scitype           = Union{Vec{Continuous},Vec{Count}},
+    prediction_type          = :deterministic,
+    orientation              = :loss,
+    reports_each_observation = true,
+    is_feature_dependent     = false,
+    supports_weights         = false,
+    docstring                = "log cosh loss; aliases: `log_cosh`.")
+
+function (log_cosh::LogCosh)(ŷ::Vec{<:Real}, y::Vec{<:Real})
+    check_dimensions(ŷ, y)
+    return log.(cosh.(ŷ-y))
+end
diff --git a/src/univariate_finite/arrays.jl b/src/univariate_finite/arrays.jl
index ff68e611..55978e09 100644
--- a/src/univariate_finite/arrays.jl
+++ b/src/univariate_finite/arrays.jl
@@ -97,9 +97,10 @@ for func in [:pdf, :logpdf]
             u::AbstractArray{UnivariateFinite{S,V,R,P},N},
             C::AbstractVector{<:Union{V, CategoricalValue{V,R}}}) where {S,V,R,P,N}
         
-            ret = Array{P,N+1}(undef, size(u)..., length(C))
+            #ret = Array{P,N+1}(undef, size(u)..., length(C))
+            ret = zeros(P, size(u)..., length(C))
             for i in eachindex(C)
-                ret[fill(:,N)...,i] = broadcast($func, u, C[i])
+                ret[fill(:,N)...,i] .= broadcast($func, u, C[i])
             end
             return ret
         end
@@ -129,8 +130,13 @@ function Base.Broadcast.broadcasted(
     cv::CategoricalValue) where {S,V,R,P,N}
 
     cv in classes(u) || _err_missing_class(cv)
-
-    return get(u.prob_given_ref, int(cv), zeros(P, size(u)))
+    
+    f() = zeros(P, size(u)) #default caller function 
+    
+    return Base.Broadcast.Broadcasted(
+    	identity,
+    	(get(f, u.prob_given_ref, int(cv)),)
+    	)
 end
         
 # pdf.(u, v)
@@ -162,41 +168,33 @@ function Base.Broadcast.broadcasted(
     raw::Union{V,AbstractArray{V,N}}) where {S,V,R,P,N}
 
     cat = transform(classes(u), raw)
-    return broadcast(pdf, u, cat)
+    return Base.Broadcast.broadcasted(pdf, u, cat)
 end
 
 # logpdf.(u::UniFinArr{S,V,R,P,N}, cv::CategoricalValue)
 # logpdf.(u::UniFinArr{S,V,R,P,N}, v::AbstractArray{<:CategoricalValue{V,R},N})
-# logpdf.(u::UniFinArr{S,V,R,P,N}, raw::V)  
-# logpdf.(u::UniFinArr{S,V,R,P,N}, raw::AbstractArray{V,N})  
+# logpdf.(u::UniFinArr{S,V,R,P,N}, raw::AbstractArray{V,N}) 
+# logpdf.(u::UniFinArr{S,V,R,P,N}, raw::V) 
 for typ in (:CategoricalValue, 
-	    :(AbstractArray{<:CategoricalValue{V,R},N}), 
+	    :(AbstractArray{<:CategoricalValue{V,R},N}),
 	    :V,
 	    :(AbstractArray{V,N}))
-    if typ == :CategoricalValue || typ == :V
-    eval(quote 
-       function Base.Broadcast.broadcasted(
-    		        ::typeof(logpdf),
-    	     	        u::UniFinArr{S,V,R,P,N},
-    	     		c::$typ) where {S,V,R,P,N}
-
-    	   # Start with the pdf array
-    	   pdf_arr = pdf.(u, c)
-    	   
-    	   # Create an uninitialized array similar to pdf_arr
-	   # this avoids mutating the initial pdf_arr  
-	   result = similar(pdf_arr)
-	   
-    	   # Take the log of each entry in-place
-    	   @simd for j in eachindex(result)
-    	       @inbounds result[j] = log(pdf_arr[j])
-    	   end
-	
-    	   return result
-        end
-    end)
-    else
+   if typ == :CategoricalValue || typ == :V
     	eval(quote
+    	function Base.Broadcast.broadcasted(
+	    		        ::typeof(logpdf),
+ 	    	     	        u::UniFinArr{S,V,R,P,N},
+	    	     		c::$typ) where {S,V,R,P,N}
+
+    	    # Start with the pdf array
+    	    # take advantage of loop fusion
+    	    result = log.(pdf.(u, c)) 
+ 	    return result
+    	end
+    	end)
+    	
+  else
+  	eval(quote
     	function Base.Broadcast.broadcasted(
 	    		        ::typeof(logpdf),
  	    	     	        u::UniFinArr{S,V,R,P,N},
@@ -213,7 +211,8 @@ for typ in (:CategoricalValue,
  	    return result
     	end
     	end)
-    end
+  end
+    
 end
 
 ## PERFORMANT BROADCASTING OF mode:
diff --git a/test/measures/continuous.jl b/test/measures/continuous.jl
index c84ce01e..a444f2fe 100644
--- a/test/measures/continuous.jl
+++ b/test/measures/continuous.jl
@@ -12,6 +12,7 @@ rng = StableRNG(666899)
     @test isapprox(mean(l1(yhat, y, w)), mae(yhat, y, w))
     @test isapprox(mean(l2(yhat, y)), 5)
     @test isapprox(mean(l2(yhat, y, w)), rms(yhat, y, w)^2)
+    @test isapprox(mean(log_cosh(yhat, y)), 1.3715546675)
 
     yhat = y .+ 1
     @test isapprox(rmsl(yhat, y),