add deepest regression estimator (#13)

Author: jbytecode

CHANGELOG.md

@@ -1,4 +1,12 @@
-# v0.10.2 (Upcoming Release)
+# v0.11.1 (Upcoming Release)
+
+
+# v0.11.0
+
+- Deepest Regression Estimator added.
+
+
+# v0.10.2 
 
 - mahalanobisSquaredBetweenPairs() return Union{Nothing, Matrix} depending on the determinant of the covariance matrix
 - mahalanobisSquaredMatrix() returns Union{Nothing, Matrix} depending on the determinant of the covariance matrix

Project.toml

@@ -1,7 +1,7 @@
 name = "LinRegOutliers"
 uuid = "6d4de0fb-32d9-4c65-aac1-cc9ed8b94b1a"
 authors = ["Mehmet Hakan Satman <mhsatman@gmail.com>", "Shreesh Adiga <16567adigashreesh@gmail.com>", "Guillermo Angeris <angeris@stanford.edu>", "Emre Akadal <emre.akadal@istanbul.edu.tr>"]
-version = "0.10.2"
+version = "0.11.0"
 
 [deps]
 Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"
@@ -14,6 +14,7 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 StatsModels = "3eaba693-59b7-5ba5-a881-562e759f1c8d"
+mrfDepth_jll = "53656f53-9700-50e7-bf9c-d3aea1338d1b"
 
 [compat]
 Clustering = "0.12.2, 0.13, 0.14, 0.15"
@@ -26,6 +27,7 @@ PrecompileTools = "1"
 Requires = "1"
 StatsModels = "0.4, 0.5, 0.6, 0.7"
 julia = "1.4"
+mrfDepth_jll = "1.0.14"
 
 [extras]
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"

README.md

@@ -38,11 +38,12 @@ A Julia package for outlier detection in linear regression.
 - Hadi (1994) Algorithm
 - Chatterjee & Mächler (1997)
 - Theil-Sen estimator for multiple regression
+- Deepest Regression Estimator
 - Summary
 
 
 ## Unimplemented Methods
-- Depth based estimators (Regression depth, deepest regression, etc.) See [#13](https://github.com/jbytecode/LinRegOutliers/issues/13) for the related issue.
+
 - Pena & Yohai (1999). See [#25](https://github.com/jbytecode/LinRegOutliers/issues/25) for the related issue.
 
 

docs/src/algorithms.md

@@ -135,6 +135,10 @@ LinRegOutliers.quantileregression
 LinRegOutliers.theilsen
 ```
 
+## Deepest Regression Estimator
+```@docs
+LinRegOutliers.deepestregression
+```
 
 
 

src/LinRegOutliers.jl

@@ -5,33 +5,33 @@ using Requires
 # After the module is loaded, we check if Plots is installed and loaded.
 # If Plots is installed and loaded, we load the corresponding modules.
 function __init__()
-    @require Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80" begin
-  
-        import .Plots: RGBX
+	@require Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80" begin
 
-        include("mveltsplot.jl")
-        include("dataimage.jl")
-        include("bchplot.jl")
+		import .Plots: RGBX
 
-        import .MVELTSPlot: mveltsplot
-        import .DataImage: dataimage
-        import .BCHPlot: bchplot
-        
-        export mveltsplot, dataimage, bchplot, RGBX
+		include("mveltsplot.jl")
+		include("dataimage.jl")
+		include("bchplot.jl")
 
-    end
+		import .MVELTSPlot: mveltsplot
+		import .DataImage: dataimage
+		import .BCHPlot: bchplot
+
+		export mveltsplot, dataimage, bchplot, RGBX
+
+	end
 end
 
 # Basis
 include("basis.jl")
 import .Basis:
-    RegressionSetting,
-    createRegressionSetting,
-    @extractRegressionSetting,
-    applyColumns,
-    find_minimum_nonzero,
-    designMatrix,
-    responseVector
+	RegressionSetting,
+	createRegressionSetting,
+	@extractRegressionSetting,
+	applyColumns,
+	find_minimum_nonzero,
+	designMatrix,
+	responseVector
 export RegressionSetting
 export createRegressionSetting
 export designMatrix
@@ -65,20 +65,20 @@ import .OrdinaryLeastSquares: OLS, ols, wls, residuals, predict, coef
 # Regression diagnostics
 include("diagnostics.jl")
 import .Diagnostics:
-    dffit,
-    dffits,
-    dfbeta,
-    dfbetas,
-    hatmatrix,
-    studentizedResiduals,
-    adjustedResiduals,
-    jacknifedS,
-    cooks,
-    cooksoutliers,
-    mahalanobisSquaredMatrix,
-    covratio,
-    hadimeasure,
-    diagnose
+	dffit,
+	dffits,
+	dfbeta,
+	dfbetas,
+	hatmatrix,
+	studentizedResiduals,
+	adjustedResiduals,
+	jacknifedS,
+	cooks,
+	cooksoutliers,
+	mahalanobisSquaredMatrix,
+	covratio,
+	hadimeasure,
+	diagnose
 
 
 # Hadi & Simonoff (1993) algorithm
@@ -205,6 +205,10 @@ import .CM97: cm97
 include("theilsen.jl")
 import .TheilSen: theilsen
 
+# Deepest Regression Estimator
+include("deepestregression.jl")
+import .DeepestRegression: deepestregression
+
 # All-in-one
 include("summary.jl")
 import .Summary: detectOutliers
@@ -267,6 +271,7 @@ export atkinson94, atkinsonstalactiteplot, generate_stalactite_plot
 export bacon
 export cm97
 export theilsen
+export deepestregression
 
 
 # Snoop-Precompile 

src/deepestregression.jl

@@ -0,0 +1,66 @@
+module DeepestRegression
+
+export deepestregression
+
+import ..Basis:
+	RegressionSetting, @extractRegressionSetting, designMatrix, responseVector
+
+using mrfDepth_jll: mrfDepth_jll
+
+
+"""
+	deepestregression(setting; maxit = 1000)
+
+Estimate Deepest Regression paramaters.
+
+
+# Arguments
+- `setting::RegressionSetting`: RegressionSetting object with a formula and dataset.
+- `maxit`: Maximum number of iterations
+
+# Description
+Estimates Deepest Regression Estimator coefficients.
+
+# References
+Van Aelst S., Rousseeuw P.J., Hubert M., Struyf A. (2002). The
+deepest regression method. Journal of Multivariate Analysis,
+81, 138-166.
+
+
+# Output
+- `betas`: Vector of regression coefficients estimated.
+"""
+function deepestregression(setting::RegressionSetting; maxit::Int = 10000)
+	X = designMatrix(setting)
+	y = responseVector(setting)
+	if all(x -> isone(x), X[:, 1])
+		X = X[:, 2:end]
+	end
+	return deepestregression(X, y, maxit = maxit)
+end
+
+function deepestregression(X::Matrix{Float64}, y::Vector{Float64}; maxit::Int = 10000)::Vector{Float64}
+	drdata = hcat(X, y)
+	n, p = size(drdata)
+	n = Int32(n)
+	p = Int32(p)
+	betas = zeros(Float64, p)
+	maxit = Int32(maxit)
+	iter = Int32(1)
+	MDEPAPPR = Int32(p)
+	ccall((:sweepmedres_, mrfDepth_jll.libmrfDepth),
+		Cint,
+		(Ref{Float64},           # X
+			Ref{Int32},          # n
+			Ref{Int32},          # np
+			Ref{Float64},        # betas
+			Ref{Cint},           # maxit 
+			Ref{Cint},           # iter
+			Ref{Cint},            # MDEPAPPR
+		), drdata, n, p, betas, maxit, iter, MDEPAPPR)
+
+	return vcat(betas[end], betas[1:(end-1)])
+end
+
+
+end # end of module DeepestRegression

src/precompile/precompile.jl

@@ -26,5 +26,6 @@ using PrecompileTools
         smr98(reg)
         ransac(reg, t = 0.8, w = 0.85)
         theilsen(reg, 2, nsamples = 10)
+        deepestregression(reg)
     end
 end

test/runtests.jl

@@ -36,3 +36,4 @@ include("testbacon2000.jl")
 include("testdataimage.jl")
 include("testtheilsen.jl")
 include("testsummary.jl")
+include("testdeepestregression.jl")

test/testdeepestregression.jl

@@ -0,0 +1,39 @@
+import LinRegOutliers: DataSets 
+
+@testset "Deepest Regression" begin 
+
+    @testset "Simple Data" begin 
+        eps = 0.1
+
+        n = 100000
+        x1 = rand(n)
+        x2 = rand(n)
+        o = ones(Float64, n)
+        e = randn(n)
+        y = 15 .+ 10 .* x1 + 5 .* x2 + e
+        X = hcat(x1, x2)
+
+        result = deepestregression(X, y)
+
+        @test isapprox(result[1], 15, atol = eps)
+        @test isapprox(result[2], 10, atol = eps)
+        @test isapprox(result[3], 5, atol = eps)
+    end 
+
+    @testset "Stackloss Data Example" begin 
+
+        eps = 0.001
+
+        setting = createRegressionSetting(
+            @formula(stackloss ~ airflow + watertemp + acidcond), 
+            DataSets.stackloss)
+        
+        result = deepestregression(setting)
+
+        @test isapprox(result[1], -35.37610619, atol = eps)
+        @test isapprox(result[2], 0.82522124, atol = eps)
+        @test isapprox(result[3], 0.44247788, atol = eps)
+        @test isapprox(result[4], -0.07964602, atol = eps)
+
+    end 
+end