mahalanobis returns Union type, code refactor

Project.toml

@@ -1,7 +1,7 @@
 name = "LinRegOutliers"
 uuid = "6d4de0fb-32d9-4c65-aac1-cc9ed8b94b1a"
 authors = ["Mehmet Hakan Satman <[email protected]>"]
-version = "0.3.12"
+version = "0.3.13"
 
 [deps]
 Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"

src/.dev/outlier.jl

@@ -61,3 +61,5 @@ include("../ransac.jl")
 
 println("ready")
 
+X = DataFrame(calls=phones.calls, year=phones.year)
+result = mve(X)

src/diagnostics.jl

@@ -319,13 +319,13 @@ Calculate Mahalanobis distances.
 Mahalanobis, Prasanta Chandra. "On the generalized distance in statistics." 
 National Institute of Science of India, 1936.
 """
-function mahalabonisSquaredMatrix(data::DataFrame; meanvector=nothing, covmatrix=nothing)::Array{Float64,2}
+function mahalabonisSquaredMatrix(data::DataFrame; meanvector=nothing, covmatrix=nothing)::Union{Nothing,Array{Float64,2}}
     datamat = convert(Matrix, data)
     return mahalabonisSquaredMatrix(datamat, meanvector=meanvector, covmatrix=covmatrix)
 end
 
 
-function mahalabonisSquaredMatrix(datamat::Matrix; meanvector=nothing, covmatrix=nothing)::Array{Float64,2}
+function mahalabonisSquaredMatrix(datamat::Matrix; meanvector=nothing, covmatrix=nothing)::Union{Nothing,Array{Float64,2}}
     if meanvector === nothing
         meanvector = applyColumns(mean, datamat)
     end
@@ -340,8 +340,7 @@ function mahalabonisSquaredMatrix(datamat::Matrix; meanvector=nothing, covmatrix
         if det(covmatrix) == 0
             @warn "singular covariance matrix, mahalanobis distances can not be calculated"
         end
-        n = size(datamat)[1]
-        return zeros(Float64, (n, n))
+        return nothing
     end
 end
 

src/mve.jl

@@ -50,7 +50,7 @@ function mve(data::DataFrame; alpha=0.05)
             meanvector = applyColumns(mean, data[hsubset, :])
             md2mat = mahalabonisSquaredMatrix(data, meanvector=meanvector, covmatrix=covmatrix)
             DJ = sqrt(sort(diag(md2mat))[h])
-            goal = (DJ / c)^p * det(covmatrix)
+            goal =  (DJ / c)^p * det(covmatrix)
         catch e
             # Possibly singularity
         end

src/ransac.jl

@@ -1,6 +1,9 @@
 """
+
     ransac(setting; t, w=0.5, m=0, k=0, d=0, confidence=0.99)
+
 Run the RANSAC (1981) algorithm for the given regression setting
+
 # Arguments
  - `setting::RegressionSetting`: RegressionSetting object with a formula and a dataset.
  - `t::Float64`: The threshold distance of a sample point to the regression hyperplane to determine if it fits the model well.
@@ -11,13 +14,13 @@ Run the RANSAC (1981) algorithm for the given regression setting
  - `confidence::Float64`: Required to determine the number of optimum iterations if k is not specified.
 
  # Examples
- ```julia-repl
+```julia-repl
 julia> df = DataFrame(y=[0,1,2,3,3,4,10], x=[0,1,2,2,3,4,2])
 julia> reg = createRegressionSetting(@formula(y ~ x), df)
 julia> ransac(reg, t=0.8, w=0.85)
 1-element Array{Int64,1}:
  7
- ```
+```
 
 # References
 Martin A. Fischler & Robert C. Bolles (June 1981). "Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography"
@@ -54,7 +57,7 @@ function ransac(setting::RegressionSetting; t::Float64, w::Float64=0.5, m::Int=0
         ols = lm(setting.formula, setting.data[sampled_indices, :])
         betas = coef(ols)
 
-        e = abs.(Y - X*betas) ./ norm([1; betas[2:end]], 2)
+        e = abs.(Y - X * betas) ./ norm([1; betas[2:end]], 2)
 
         iteration_inlier_indices = filter(i -> e[i] < t, 1:n)
         inliers_count = length(iteration_inlier_indices)

src/satman2013.jl

@@ -48,7 +48,12 @@ function satman2013(setting::RegressionSetting)
     end
 
     medians = applyColumns(median, X0)
-    md2 = diag(mahalabonisSquaredMatrix(X0, meanvector = medians, covmatrix = covmat))
+    mhs = mahalabonisSquaredMatrix(X0, meanvector=medians, covmatrix=covmat)
+    if mhs isa Nothing
+        md2 = zeros(Float64, n)
+    else
+        md2 = diag(mhs)
+    end
     md = sqrt.(md2)
 
     sorted_indices = sortperm(md)

src/satman2015.jl

@@ -233,7 +233,12 @@ function satman2015(setting::RegressionSetting)
 
     meanvector = applyColumns(mean, X[basic_subset_indices,:])
     covmat = cov(X[basic_subset_indices,:])
-    md2 = diag(mahalabonisSquaredMatrix(X, meanvector = meanvector, covmatrix = covmat))
+    mhs = mahalabonisSquaredMatrix(X, meanvector = meanvector, covmatrix = covmat)
+    if mhs isa Nothing
+        md2 = zeros(Float64, n)
+    else
+        md2 = diag(mhs)
+    end
     md = sqrt.(md2)
     sorted_indices = sortperm(md)
     best_h_indices = sorted_indices[1:h]