Merge pull request #244 from alan-turing-institute/dev

For a 0.12.7 release

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJBase"
 uuid = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.12.6"
+version = "0.12.7"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
@@ -38,7 +38,7 @@ HTTP = "^0.8"
 InvertedIndices = "^1"
 JLSO = "^2.1,^2.2"
 JSON = "^0.21"
-LossFunctions = "^0.5"
+LossFunctions = "0.5, 0.6"
 MLJModelInterface = "^0.2"
 MLJScientificTypes = "^0.1,^0.2"
 Missings = "^0.4"

src/measures/loss_functions_interface.jl

@@ -1,7 +1,7 @@
 # implementation of MLJ measure interface for LossFunctions.jl
 
 # Supervised Loss -- measure traits
-
+const LSF = LossFunctions
 is_measure_type(::Type{<:SupervisedLoss})          = true
 orientation(::Type{<:SupervisedLoss})              = :loss
 reports_each_observation(::Type{<:SupervisedLoss}) = true
@@ -18,12 +18,12 @@ MMI.target_scitype(::Type{<:DistanceLoss})  = AbstractArray{<:Continuous}
 
 function value(measure::DistanceLoss, yhat, X, y, ::Nothing,
                 ::Val{false}, ::Val{true})
-    return measure(yhat, y)
+    return LSF.value(measure, yhat, y)
 end
 
 function value(measure::DistanceLoss, yhat, X, y, w,
                 ::Val{false}, ::Val{true})
-    return w .* measure(yhat, y) ./ (sum(w)/length(y))
+    return w .* value(measure, yhat, X, y, nothing) ./ (sum(w)/length(y))
 end
 
 ## MARGIN BASED LOSS FUNCTIONS
@@ -42,7 +42,7 @@ function value(measure::MarginLoss, yhat, X, y, ::Nothing,
                 ::Val{false}, ::Val{true})
     check_pools(yhat, y)
     probs_of_observed = broadcast(pdf, yhat, y)
-    return broadcast(measure, _scale.(probs_of_observed), 1)
+    return (LSF.value).(measure, _scale.(probs_of_observed), 1)
 end
 
 function value(measure::MarginLoss, yhat, X, y, w,

test/measures/loss_functions_interface.jl

@@ -29,7 +29,7 @@ seed!(1234)
     for m in [ZeroOneLoss(), L1HingeLoss(), L2HingeLoss(), LogitMarginLoss(),
               ModifiedHuberLoss(), PerceptronLoss(), SmoothedL1HingeLoss(0.9),
               L2MarginLoss(), ExpLoss(), SigmoidLoss(), DWDMarginLoss(0.9)]
-        @test MLJBase.value(m, yhat, X, y, nothing) ≈ m(yhatm, ym)
+        @test MLJBase.value(m, yhat, X, y, nothing) ≈ LossFunctions.value(m, yhatm, ym)
         @test mean(MLJBase.value(m, yhat, X, y, w)) ≈
                 LossFunctions.value(m, yhatm, ym, AggMode.WeightedMean(w))
     end
@@ -47,7 +47,7 @@ end
               HuberLoss(0.9), EpsilonInsLoss(0.9), L1EpsilonInsLoss(0.9),
               L2EpsilonInsLoss(0.9), LogitDistLoss(), QuantileLoss(0.7)]
 
-        @test MLJBase.value(m, yhat, X, y, nothing) ≈ m(yhat, y)
+        @test MLJBase.value(m, yhat, X, y, nothing) ≈ LossFunctions.value(m, yhat, y)
         @test mean(MLJBase.value(m, yhat, X, y, w)) ≈
                 LossFunctions.value(m, yhat, y, AggMode.WeightedMean(w))
     end