Do not extended Statistics.cor/Statistics.cov (#148)

Using the same functions as the `Statistics` ones is a pun, but the functions do
different things, in particular they return scalar values, while the method
defined here return matrices and make it impossible to do the same thing as the
`Statistics` methods.

Also, adding the `Statistics` package as dependency with the sole purpose of
extending methods of functions, which could have been done with an extension,
increases loading time by a factor of 40.

Project.toml

@@ -9,7 +9,6 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 RecipesBase = "3cdcf5f2-1ef4-517c-9805-6587b60abb01"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
-Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [weakdeps]
 Juno = "e5e0dc1b-0480-54bc-9374-aad01c23163d"

docs/src/examples.md

@@ -450,8 +450,7 @@ Calculating the Covariance and Correlation Matrices
 ---------------------------------------------------
 
 Calculate the covariance and correlation matrices of multiple `Measurement`s
-with the functions [`cov(::AbstractVector{<:Measurement})`](@ref) and
-[`cor(::AbstractVector{<:Measurement})`](@ref):
+with the functions [`Measurements.cov`](@ref) and [`Measurements.cor`](@ref):
 
 ```@repl
 using Measurements

docs/src/usage.md

@@ -210,8 +210,8 @@ Calculating the Covariance and Correlation Matrices
 ---------------------------------------------------
 
 ```@docs
-Measurements.Statistics.cov(::AbstractVector{<:Measurement})
-Measurements.Statistics.cor(::AbstractVector{<:Measurement})
+Measurements.cov
+Measurements.cor
 ```
 
 The [covariance matrix](https://en.wikipedia.org/wiki/Covariance_matrix) of

src/Measurements.jl

@@ -23,14 +23,9 @@ module Measurements
 
 # Calculus is used to calculate numerical derivatives in "@uncertain" macro.
 using Calculus
-using Statistics
-
-import Statistics: cor, cov
 
 # Functions provided by this package and exposed to users
 export Measurement, measurement, ±
-# Re-export from Statistics
-export cor, cov
 
 # Define the "Derivatives" type, used inside "Measurement" type.  This should be
 # a lightweight and immutable dictionary.

src/utils.jl

@@ -107,11 +107,11 @@ function uncertainty_components(x::Measurement{T}) where {T<:AbstractFloat}
 end
 
 """
-    cov(x::AbstractVector{<:Measurement})
+    Measurements.cov(x::AbstractVector{<:Measurement})
 
 Returns the covariance matrix of a vector of correlated `Measurement`s.
 """
-function Statistics.cov(x::AbstractVector{Measurement{T}}) where T
+function cov(x::AbstractVector{Measurement{T}}) where T
     S = length(x)
     covariance_matrix = zeros(T, (S, S))
 
@@ -126,11 +126,11 @@ function Statistics.cov(x::AbstractVector{Measurement{T}}) where T
 end
 
 """
-    cor(x::AbstractVector{<:Measurement})
+    Measurements.cor(x::AbstractVector{<:Measurement})
 
 Returns the correlation matrix of a vector of correlated `Measurement`s.
 """
-function Statistics.cor(x::AbstractVector{<:Measurement})
+function cor(x::AbstractVector{<:Measurement})
     covariance_matrix = cov(x)
     standard_deviations = sqrt.(diag(covariance_matrix))
     return covariance_matrix ./ standard_deviations ./ standard_deviations'

test/runtests.jl

@@ -103,30 +103,30 @@ end
     x = measurement(1.0, 0.1)
     y = -2x + 10
     z = -3x
-    @test @inferred(cov([x, y, z])) ≈ [0.01 -0.02 -0.03; -0.02 0.04 0.06; -0.03 0.06 0.09]
+    @test @inferred(Measurements.cov([x, y, z])) ≈ [0.01 -0.02 -0.03; -0.02 0.04 0.06; -0.03 0.06 0.09]
 
     u = measurement(1, 0.05)
     v = measurement(10, 0.1)
     w = u + 2v
-    @test @inferred(cov([u, v, w])) ≈ [0.0025 0.0 0.0025; 0.0 0.01 0.02; 0.0025 0.02 0.0425]
+    @test @inferred(Measurements.cov([u, v, w])) ≈ [0.0025 0.0 0.0025; 0.0 0.01 0.02; 0.0025 0.02 0.0425]
 end
 
 @testset "Correlation matrix" begin
     x = measurement(1.0, 0.1)
     y = -2x + 10
     z = -3x
-    @test @inferred(cor([x, y, z])) ≈ [1.0 -1.0 -1.0; -1.0 1.0 1.0; -1.0 1.0 1.0]
+    @test @inferred(Measurements.cor([x, y, z])) ≈ [1.0 -1.0 -1.0; -1.0 1.0 1.0; -1.0 1.0 1.0]
 
     u = measurement(1, 0.05)
     v = measurement(10, 0.1)
     w = u + 2v
-    @test @inferred(cor([u, v, w])) ≈ [1.0 0.0 0.24253562503633297; 0.0 1.0 0.9701425001453319; 0.24253562503633297 0.9701425001453319 1.0] atol=1e-15
+    @test @inferred(Measurements.cor([u, v, w])) ≈ [1.0 0.0 0.24253562503633297; 0.0 1.0 0.9701425001453319; 0.24253562503633297 0.9701425001453319 1.0] atol=1e-15
 end
 
 @testset "Correlated values" begin
     @testset "Simple" begin
         u_in = measurement(1, 0.1)
-        cov_matrix = @inferred cov([u_in])
+        cov_matrix = @inferred Measurements.cov([u_in])
 
         u_out, = @inferred Measurements.correlated_values([1], cov_matrix)
         @test 2u_in ≈ 2u_out
@@ -139,7 +139,7 @@ end
 
         xyz_in = [x_in, y_in, z_in]
 
-        cov_matrix = cov([x_in, y_in, z_in])
+        cov_matrix = Measurements.cov([x_in, y_in, z_in])
 
         @test Measurements.uncertainty.([x_in, y_in, z_in]) .^ 2 ≈ diag(cov_matrix)
 
@@ -155,7 +155,7 @@ end
         v_in = measurement(10, 0.1)
         w_in = u_in + 2v_in
 
-        cov_matrix = cov([u_in, v_in, w_in])
+        cov_matrix = Measurements.cov([u_in, v_in, w_in])
 
         (u_out, v_out, w_out) = Measurements.correlated_values(Measurements.value.([u_in, v_in, w_in]), cov_matrix)
 
@@ -165,7 +165,7 @@ end
         @test w_out ≈ u_out + 2v_out
         @test 0 ± 0 ≈ u_out + 2v_out - w_out atol=2e-8
 
-        corr_matrix = cor([u_out, v_out, w_out])
+        corr_matrix = Measurements.cor([u_out, v_out, w_out])
         @test corr_matrix[1,1] ≈ 1
         @test corr_matrix[2,3] ≈ 2Measurements.uncertainty(v_out)/Measurements.uncertainty(w_out)
     end
@@ -183,7 +183,7 @@ end
         x_in = measurement(1, 0)
         y_in = measurement(2, 0)
         z_in = measurement(3, 5)
-        cov_matrix = cov([x_in, y_in, z_in])
+        cov_matrix = Measurements.cov([x_in, y_in, z_in])
         nom_values = Measurements.value.([x_in, y_in, z_in])
         variables = Measurements.correlated_values(nom_values, cov_matrix)
 
@@ -194,7 +194,7 @@ end
             @test Measurements.uncertainty(variable)^2 ≈ variance
         end
 
-        @test cov_matrix ≈ cov(variables)
+        @test cov_matrix ≈ Measurements.cov(variables)
         @test [x_in, y_in, z_in] ≈ variables
     end
 end