Merge pull request #122 from juliohm/misclassification

Add MisclassLoss as a generalization of ZeroOneLoss

src/LossFunctions.jl

@@ -31,12 +31,15 @@ import LearnBase:
 
 export
 
+    value,
+    value!,
     deriv2!,
     value_fun,
     deriv_fun,
     deriv2_fun,
     value_deriv_fun,
 
+    ZeroOneLoss,
     LogitMarginLoss,
     PerceptronLoss,
     HingeLoss,
@@ -61,6 +64,7 @@ export
     QuantileLoss,
     PinballLoss,
 
+    MisclassLoss,
     PoissonLoss,
     LogitProbLoss,
     CrossEntropyLoss,

src/supervised/other.jl

@@ -1,3 +1,36 @@
+@doc doc"""
+    MisclassLoss <: SupervisedLoss
+
+Misclassification loss that assigns `1` for misclassified
+examples and `0` otherwise. It is a generalization of
+`ZeroOneLoss` for more than two classes.
+"""
+struct MisclassLoss <: SupervisedLoss end
+
+agreement(target, output) = target == output
+
+value(::MisclassLoss, agreement::Bool) = agreement ? 0 : 1
+deriv(::MisclassLoss, agreement::Bool) = 0
+deriv2(::MisclassLoss, agreement::Bool) = 0
+value_deriv(::MisclassLoss, agreement::Bool) = agreement ? (0, 0) : (1, 0)
+
+value(loss::MisclassLoss, target::Number, output::Number) = value(loss, agreement(target, output))
+deriv(loss::MisclassLoss, target::Number, output::Number) = deriv(loss, agreement(target, output))
+deriv2(loss::MisclassLoss, target::Number, output::Number) = deriv2(loss, agreement(target, output))
+
+isminimizable(::MisclassLoss) = false
+isdifferentiable(::MisclassLoss) = false
+isdifferentiable(::MisclassLoss, at) = at != 0
+istwicedifferentiable(::MisclassLoss) = false
+istwicedifferentiable(::MisclassLoss, at) = at != 0
+isnemitski(::MisclassLoss) = false
+islipschitzcont(::MisclassLoss) = false
+isconvex(::MisclassLoss) = false
+isclasscalibrated(::MisclassLoss) = false
+isclipable(::MisclassLoss) = false
+
+# ===============================================================
+
 @doc doc"""
     PoissonLoss <: SupervisedLoss
 

test/runtests.jl

@@ -35,6 +35,10 @@ margin_losses = [
     DWDMarginLoss(1), DWDMarginLoss(2)
 ]
 
+other_losses = [
+    MisclassLoss(), PoissonLoss(), CrossEntropyLoss()
+]
+
 for t in tests
     @testset "$t" begin
         include(t)

test/tst_loss.jl

@@ -483,6 +483,9 @@ end
 
 
 @testset "Test other loss against reference function" begin
+    _misclassloss(y, t) = y == t ? 0 : 1
+    test_value(MisclassLoss(), _misclassloss, 1:10, vcat(1:5,7:11))
+
     _crossentropyloss(y, t) = -y*log(t) - (1-y)*log(1-t)
     test_value(CrossEntropyLoss(), _crossentropyloss, 0:0.01:1, 0.01:0.01:0.99)