Merge pull request #259 from JuliaML/parameters

Implement the `parameters` function for some transforms

Author: juliohm

Project.toml

@@ -38,6 +38,6 @@ Statistics = "1.9"
 StatsBase = "0.33, 0.34"
 Tables = "1.6"
 Transducers = "0.4"
-TransformsBase = "1.3"
+TransformsBase = "1.4"
 Unitful = "1.17"
 julia = "1.9"

src/TableTransforms.jl

@@ -27,7 +27,7 @@ using Transducers: tcollect
 using NelderMead: optimise
 
 import Distributions: quantile, cdf
-import TransformsBase: assertions, isrevertible, isinvertible
+import TransformsBase: assertions, parameters, isrevertible, isinvertible
 import TransformsBase: apply, revert, reapply, preprocess, inverse
 
 include("tabletraits.jl")

src/transforms/coalesce.jl

@@ -42,6 +42,8 @@ Coalesce(; value) = Coalesce(AllSelector(), value)
 Coalesce(cols; value) = Coalesce(selector(cols), value)
 Coalesce(cols::C...; value) where {C<:Column} = Coalesce(selector(cols), value)
 
+parameters(transform::Coalesce) = (; value=transform.value)
+
 isrevertible(::Type{<:Coalesce}) = false
 
 colcache(::Coalesce, x) = nothing

src/transforms/dropextrema.jl

@@ -50,6 +50,8 @@ DropExtrema(; low=0.25, high=0.75) = DropExtrema(AllSelector(), low, high)
 DropExtrema(cols; low=0.25, high=0.75) = DropExtrema(selector(cols), low, high)
 DropExtrema(cols::C...; low=0.25, high=0.75) where {C<:Column} = DropExtrema(selector(cols), low, high)
 
+parameters(transform::DropExtrema) = (low=transform.low, high=transform.high)
+
 isrevertible(::Type{<:DropExtrema}) = false
 
 function preprocess(transform::DropExtrema, feat)

src/transforms/eigenanalysis.jl

@@ -54,6 +54,8 @@ EigenAnalysis(proj; maxdim=nothing, pratio=1.0) = EigenAnalysis(proj, maxdim, pr
 
 assertions(::EigenAnalysis) = [scitypeassert(Continuous)]
 
+parameters(transform::EigenAnalysis) = (proj=transform.proj, maxdim=transform.maxdim, pratio=transform.pratio)
+
 isrevertible(::Type{EigenAnalysis}) = true
 
 function applyfeat(transform::EigenAnalysis, feat, prep)

src/transforms/indicator.jl

@@ -48,6 +48,8 @@ Indicator(col::Column; k=10, scale=:quantile, categ=false) = Indicator(selector(
 
 assertions(transform::Indicator) = [scitypeassert(Continuous, transform.selector)]
 
+parameters(transform::Indicator) = (k=transform.k, scale=transform.scale)
+
 isrevertible(::Type{<:Indicator}) = true
 
 function _intervals(transform::Indicator, x)

src/transforms/projectionpursuit.jl

@@ -49,6 +49,8 @@ ProjectionPursuit(; tol=1e-6, maxiter=100, deg=5, perc=0.9, n=100, rng=Random.GL
 
 assertions(::ProjectionPursuit) = [scitypeassert(Continuous)]
 
+parameters(transform::ProjectionPursuit) = (tol=transform.tol, deg=transform.deg, perc=transform.perc, n=transform.n)
+
 isrevertible(::Type{<:ProjectionPursuit}) = true
 
 # transforms a row of random variables into a convex combination 

src/transforms/quantile.jl

@@ -41,6 +41,8 @@ Quantile(cols::C...; dist=Normal()) where {C<:Column} = Quantile(selector(cols),
 
 assertions(transform::Quantile) = [scitypeassert(Continuous, transform.selector)]
 
+parameters(transform::Quantile) = (; dist=transform.dist)
+
 isrevertible(::Type{<:Quantile}) = true
 
 colcache(::Quantile, x) = EmpiricalDistribution(x)

src/transforms/remainder.jl

@@ -22,6 +22,8 @@ isrevertible(::Type{<:Remainder}) = true
 
 assertions(::Remainder) = [scitypeassert(Continuous)]
 
+parameters(transform::Remainder) = (; total=transform.total)
+
 function applyfeat(transform::Remainder, feat, prep)
   cols = Tables.columns(feat)
   names = Tables.columnnames(cols) |> collect

src/transforms/scale.jl

@@ -50,6 +50,8 @@ Scale(cols::C...; low=0.25, high=0.75) where {C<:Column} = Scale(selector(cols),
 
 assertions(transform::Scale) = [scitypeassert(Continuous, transform.selector)]
 
+parameters(transform::Scale) = (low=transform.low, high=transform.high)
+
 isrevertible(::Type{<:Scale}) = true
 
 function colcache(transform::Scale, x)

test/transforms/coalesce.jl

@@ -1,6 +1,8 @@
 @testset "Coalesce" begin
   @test !isrevertible(Coalesce(value=0))
 
+  @test TT.parameters(Coalesce(value=0)) == (; value=0)
+
   a = [3, 2, missing, 4, 5, 3]
   b = [missing, 4, 4, 5, 8, 5]
   c = [1, 1, 6, 2, 4, missing]

test/transforms/dropextrema.jl

@@ -1,6 +1,8 @@
 @testset "DropExtrema" begin
   @test !isrevertible(DropExtrema(:a))
 
+  @test TT.parameters(DropExtrema(:a)) == (low=0.25, high=0.75)
+
   a = [6.9, 9.0, 7.8, 0.0, 5.1, 4.8, 1.1, 8.0, 5.4, 7.9]
   b = [7.7, 4.2, 6.3, 1.4, 4.4, 0.5, 3.0, 6.1, 1.9, 1.5]
   c = [6.1, 7.7, 5.7, 2.8, 2.8, 6.7, 8.4, 5.0, 8.9, 1.0]

test/transforms/eigenanalysis.jl

@@ -1,4 +1,6 @@
 @testset "EigenAnalysis" begin
+  @test TT.parameters(EigenAnalysis(:V)) == (proj=:V, maxdim=nothing, pratio=1.0)
+
   # PCA test
   x = rand(Normal(0, 10), 1500)
   y = x + rand(Normal(0, 2), 1500)

test/transforms/indicator.jl

@@ -1,4 +1,6 @@
 @testset "Indicator" begin
+  @test TT.parameters(Indicator(:a)) == (k=10, scale=:quantile)
+
   a = [5.8, 6.4, 6.4, 9.8, 7.6, 8.2, 4.5, 2.5, 1.7, 2.3]
   b = [8.4, 1.4, 7.2, 1.8, 9.4, 1.0, 2.0, 5.2, 9.4, 6.2]
   c = [4.1, 5.6, 7.1, 9.1, 5.9, 9.5, 5.7, 9.0, 6.6, 9.9]

test/transforms/projectionpursuit.jl

@@ -1,4 +1,6 @@
 @testset "ProjectionPursuit" begin
+  @test TT.parameters(ProjectionPursuit()) == (tol=1.0e-6, deg=5, perc=0.9, n=100)
+
   rng = MersenneTwister(42)
   N = 10_000
   a = [2randn(rng, N ÷ 2) .+ 6; randn(rng, N ÷ 2)]

test/transforms/quantile.jl

@@ -1,4 +1,6 @@
 @testset "Quantile" begin
+  @test TT.parameters(Quantile()) == (; dist=Normal())
+
   t = Table(z=rand(100))
   T = Quantile()
   n, c = apply(T, t)

test/transforms/remainder.jl

@@ -1,6 +1,8 @@
 @testset "Remainder" begin
   @test isrevertible(Remainder())
 
+  @test TT.parameters(Remainder()) == (; total=nothing)
+
   a = [2.0, 66.0, 0.0]
   b = [4.0, 22.0, 2.0]
   c = [4.0, 12.0, 98.0]

test/transforms/scale.jl

@@ -1,4 +1,6 @@
 @testset "Scale" begin
+  @test TT.parameters(Scale(:a)) == (low=0.25, high=0.75)
+
   # constant column
   x = fill(3.0, 10)
   y = rand(10)