Log Transforms

Project.toml

@@ -1,7 +1,7 @@
 name = "FeatureTransforms"
 uuid = "8fd68953-04b8-4117-ac19-158bf6de9782"
 authors = ["Invenia Technical Computing Corporation"]
-version = "0.3.10"
+version = "0.3.11"
 
 [deps]
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"

docs/src/api.md

@@ -15,7 +15,9 @@ Power
 Periodic
 MeanStdScaling
 IdentityScaling
+InverseHyperbolicSine
 LinearCombination
+LogTransform
 OneHotEncoding
 ```
 

src/FeatureTransforms.jl

@@ -8,6 +8,7 @@ using Tables
 export Transform, transform, transform!
 export HoD, LinearCombination, OneHotEncoding, Periodic, Power
 export AbstractScaling, IdentityScaling, MeanStdScaling
+export LogTransform, InverseHyperbolicSine
 
 include("utils.jl")
 include("traits.jl")
@@ -16,6 +17,7 @@ include("apply.jl")
 
 # Transform implementations
 include("linear_combination.jl")
+include("log.jl")
 include("one_hot_encoding.jl")
 include("periodic.jl")
 include("power.jl")

src/log.jl

@@ -0,0 +1,33 @@
+"""
+    LogTransform <: Transform
+
+Logarithmically transform the data through: sign(x) * log(|x| + 1).
+
+This allows transformations of all real numbers, not just positive ones.
+"""
+struct LogTransform <: Transform end
+cardinality(::LogTransform) = OneToOne()
+
+_logtransform(x) = sign(x) * log(abs(x) + one(x))
+_invlogtransform(x) = sign(x) * (exp(sign(x) * x) - one(x))
+
+function _apply(A::AbstractArray, transform::LogTransform; inverse=false, kwargs...)
+    inverse && return _invlogtransform.(A)
+    return _logtransform.(A)
+end
+
+
+"""
+    InverseHyperbolicSine <: Transform
+
+Logarithmically transform the data through: log(x + √(x² + 1)).
+
+This is the inverse hyperbolic sine. 
+"""
+struct InverseHyperbolicSine <: Transform end
+cardinality(::InverseHyperbolicSine) = OneToOne()
+
+function _apply(A::AbstractArray, transform::InverseHyperbolicSine; inverse=false, kwargs...)
+    inverse && return sinh.(A)
+    return asinh.(A)
+end      

test/log.jl

@@ -0,0 +1,46 @@
+@testset "log" begin
+
+    V1 = [1; -2; 3; 4; 0; -6]
+    V2 = -V1
+    M = [1 1 0.5; 0.0 1.0 2.0]
+
+    @testset "LogTransform" begin
+
+        logV1 = [log(2); -log(3); log(4); log(5); log(1); -log(7)]
+        logV2 = -logV1
+        logM = [log(2) log(2) log(1.5); log(1) log(2) log(3)]
+
+        @testset "simple" for x in (V1, V2, M)
+            transform = LogTransform()
+            @test cardinality(transform) == OneToOne()
+            @test transform isa Transform
+        end
+
+        @testset "Apply" for (x, y) in ((V1, logV1), (V2, logV2), (M, logM))
+            f = LogTransform()
+            transformed = FeatureTransforms.apply(x, f)
+            @test transformed ≈ y atol=1e-5
+            @test FeatureTransforms.apply(transformed, f; inverse=true) ≈ x atol=1e-5
+        end
+    end
+
+    @testset "InverseHyperbolicSine" begin
+
+        logV1 = asinh.(V1)
+        logV2 = asinh.(V2)
+        logM = asinh.(M)
+
+        @testset "simple" for x in (V1, V2, M)
+            transform = InverseHyperbolicSine()
+            @test cardinality(transform) == OneToOne()
+            @test transform isa Transform
+        end
+
+        @testset "Apply" for (x, y) in ((V1, logV1), (V2, logV2), (M, logM))
+            f = InverseHyperbolicSine()
+            transformed = FeatureTransforms.apply(x, f)
+            @test transformed ≈ y atol=1e-5
+            @test FeatureTransforms.apply(transformed, f; inverse=true) ≈ x atol=1e-5
+        end
+    end
+end

test/runtests.jl

@@ -17,6 +17,7 @@ using TimeZones
     Sys.WORD_SIZE == 64 && v"1.6" <= VERSION < v"1.7" && doctest(FeatureTransforms)
 
     include("linear_combination.jl")
+    include("log.jl")
     include("one_hot_encoding.jl")
     include("periodic.jl")
     include("power.jl")