Add testsets (#83)

add testsets
JuliaStats · Dec 21, 2018 · 26dd9d1 · 26dd9d1
1 parent 924cb34
commit 26dd9d1
Show file tree

Hide file tree

Showing 10 changed files with 935 additions and 902 deletions.
diff --git a/test/cmds.jl b/test/cmds.jl
@@ -2,58 +2,62 @@ using MultivariateStats
 using LinearAlgebra
 using Test
 
-## testing data
+@testset "MDS" begin
 
-function pwdists(X)
-    S = sum(abs2, X, dims=1)
-    D2 = S .+ S' - 2 * (X'X)
-    D2[diagind(D2)] .= 0.0
-    return sqrt.(D2)
-end
+    ## testing data
 
-X0 = randn(3, 5)
-G0 = X0'X0
-D0 = pwdists(X0)
+    function pwdists(X)
+        S = sum(abs2, X, dims=1)
+        D2 = S .+ S' - 2 * (X'X)
+        D2[diagind(D2)] .= 0.0
+        return sqrt.(D2)
+    end
 
-## conversion between dmat and gram
+    X0 = randn(3, 5)
+    G0 = X0'X0
+    D0 = pwdists(X0)
 
-@assert issymmetric(D0)
-@assert issymmetric(G0)
+    ## conversion between dmat and gram
 
-D = gram2dmat(G0)
-@test issymmetric(D)
-@test D ≈ D0
+    @assert issymmetric(D0)
+    @assert issymmetric(G0)
 
-G = dmat2gram(D0)
-@test issymmetric(G)
-@test gram2dmat(G) ≈ D0
+    D = gram2dmat(G0)
+    @test issymmetric(D)
+    @test D ≈ D0
 
-## classical MDS
+    G = dmat2gram(D0)
+    @test issymmetric(G)
+    @test gram2dmat(G) ≈ D0
 
-X = classical_mds(D0, 3)
-@test pwdists(X) ≈ D0
+    ## classical MDS
 
-#Test MDS embeddings in dimensions >= number of points
-@test classical_mds([0 1; 1 0], 2, dowarn=false) == [-0.5 0.5; 0 0]
-@test classical_mds([0 1; 1 0], 3, dowarn=false) == [-0.5 0.5; 0 0; 0 0]
+    X = classical_mds(D0, 3)
+    @test pwdists(X) ≈ D0
 
+    #Test MDS embeddings in dimensions >= number of points
+    @test classical_mds([0 1; 1 0], 2, dowarn=false) == [-0.5 0.5; 0 0]
+    @test classical_mds([0 1; 1 0], 3, dowarn=false) == [-0.5 0.5; 0 0; 0 0]
 
-#10 - dmat2gram produces negative definite matrix
-let D = [1.0 0.5 0.3181818181818182 0.38095238095238093 0.6111111111111112 0.36363636363636365 0.3333333333333333 0.4444444444444444 0.45454545454545453 0.38095238095238093
-        0.5 1.0 0.3333333333333333 0.4166666666666667 0.32 0.37037037037037035 0.35 0.38461538461538464 0.38461538461538464 0.4
-        0.3181818181818182 0.3333333333333333 1.0 0.38095238095238093 0.125 0.22727272727272727 0.47058823529411764 0.35294117647058826 0.2608695652173913 0.21052631578947367
-        0.38095238095238093 0.4166666666666667 0.38095238095238093 1.0 0.17391304347826086 0.2962962962962963 0.38095238095238093 0.45454545454545453 0.3076923076923077 0.5454545454545454
-        0.6111111111111112 0.32 0.125 0.17391304347826086 1.0 0.5 0.15 0.391304347826087 0.38095238095238093 0.3333333333333333
-        0.36363636363636365 0.37037037037037035 0.22727272727272727 0.2962962962962963 0.5 1.0 0.2857142857142857 0.44 0.4 0.3181818181818182
-        0.3333333333333333 0.35 0.47058823529411764 0.38095238095238093 0.15 0.2857142857142857 1.0 0.42105263157894735 0.45 0.23529411764705882
-        0.4444444444444444 0.38461538461538464 0.35294117647058826 0.45454545454545453 0.391304347826087 0.44 0.42105263157894735 1.0 0.5416666666666666 0.5555555555555556
-        0.45454545454545453 0.38461538461538464 0.2608695652173913 0.3076923076923077 0.38095238095238093 0.4 0.45 0.5416666666666666 1.0 0.4
-        0.38095238095238093 0.4 0.21052631578947367 0.5454545454545454 0.3333333333333333 0.3181818181818182 0.23529411764705882 0.5555555555555556 0.4 1.0],
-    Xt = [-0.27529104101488666 0.006134513718202863 0.33298809606740326 0.2608994458893664 -0.46185275796909575 -0.23734315039370618 0.29972782027671513 0.03827901455843394 -0.04096713097883363 0.07742518984640051
-        -0.08177061420820278 -0.0044504235228030225 -0.3271919093638943 0.28206254638779243 -0.0954706915166714 -0.07137742126520012 -0.30754933764853587 0.18582658369448027 -0.03715307349750036 0.45707434094053534]',
-    M = classical_mds(D, 2)'
-    @test M ≈ Xt .* cis.(angle.(sum(conj.(Xt) .* M, dims=1)))
-end
 
-#10 - test degenerate problem
-@test classical_mds(zeros(10, 10), 3, dowarn=false) == zeros(3, 10)
+    #10 - dmat2gram produces negative definite matrix
+    let D = [1.0 0.5 0.3181818181818182 0.38095238095238093 0.6111111111111112 0.36363636363636365 0.3333333333333333 0.4444444444444444 0.45454545454545453 0.38095238095238093
+            0.5 1.0 0.3333333333333333 0.4166666666666667 0.32 0.37037037037037035 0.35 0.38461538461538464 0.38461538461538464 0.4
+            0.3181818181818182 0.3333333333333333 1.0 0.38095238095238093 0.125 0.22727272727272727 0.47058823529411764 0.35294117647058826 0.2608695652173913 0.21052631578947367
+            0.38095238095238093 0.4166666666666667 0.38095238095238093 1.0 0.17391304347826086 0.2962962962962963 0.38095238095238093 0.45454545454545453 0.3076923076923077 0.5454545454545454
+            0.6111111111111112 0.32 0.125 0.17391304347826086 1.0 0.5 0.15 0.391304347826087 0.38095238095238093 0.3333333333333333
+            0.36363636363636365 0.37037037037037035 0.22727272727272727 0.2962962962962963 0.5 1.0 0.2857142857142857 0.44 0.4 0.3181818181818182
+            0.3333333333333333 0.35 0.47058823529411764 0.38095238095238093 0.15 0.2857142857142857 1.0 0.42105263157894735 0.45 0.23529411764705882
+            0.4444444444444444 0.38461538461538464 0.35294117647058826 0.45454545454545453 0.391304347826087 0.44 0.42105263157894735 1.0 0.5416666666666666 0.5555555555555556
+            0.45454545454545453 0.38461538461538464 0.2608695652173913 0.3076923076923077 0.38095238095238093 0.4 0.45 0.5416666666666666 1.0 0.4
+            0.38095238095238093 0.4 0.21052631578947367 0.5454545454545454 0.3333333333333333 0.3181818181818182 0.23529411764705882 0.5555555555555556 0.4 1.0],
+        Xt = [-0.27529104101488666 0.006134513718202863 0.33298809606740326 0.2608994458893664 -0.46185275796909575 -0.23734315039370618 0.29972782027671513 0.03827901455843394 -0.04096713097883363 0.07742518984640051
+            -0.08177061420820278 -0.0044504235228030225 -0.3271919093638943 0.28206254638779243 -0.0954706915166714 -0.07137742126520012 -0.30754933764853587 0.18582658369448027 -0.03715307349750036 0.45707434094053534]',
+        M = classical_mds(D, 2)'
+        @test M ≈ Xt .* cis.(angle.(sum(conj.(Xt) .* M, dims=1)))
+    end
+
+    #10 - test degenerate problem
+    @test classical_mds(zeros(10, 10), 3, dowarn=false) == zeros(3, 10)
+
+end
diff --git a/test/fa.jl b/test/fa.jl
@@ -4,106 +4,110 @@ using Test
 import Statistics: mean, cov, var
 import Random
 
-Random.seed!(34568)
+@testset "Factor Analysis" begin
 
-## FA with zero mean
+    Random.seed!(34568)
 
-X = randn(5, 10)
-Y = randn(3, 10)
+    ## FA with zero mean
 
-W = qr(randn(5, 5)).Q[:, 1:3]
-Ψ = fill(0.1, 5)
-M = FactorAnalysis(Float64[], W, Ψ)
+    X = randn(5, 10)
+    Y = randn(3, 10)
 
-@test indim(M) == 5
-@test outdim(M) == 3
-@test mean(M) == zeros(5)
-@test loadings(M) == W
-@test var(M) == Ψ
+    W = qr(randn(5, 5)).Q[:, 1:3]
+    Ψ = fill(0.1, 5)
+    M = FactorAnalysis(Float64[], W, Ψ)
 
-T = inv(I+W'*diagm(0 => 1 ./ var(M))*W)*W'*diagm(0 => 1 ./ var(M))
-@test transform(M, X[:,1]) ≈ T * X[:,1]
-@test transform(M, X) ≈ T * X
+    @test indim(M) == 5
+    @test outdim(M) == 3
+    @test mean(M) == zeros(5)
+    @test loadings(M) == W
+    @test var(M) == Ψ
 
-R = cov(M)*W*inv(W'W)
-@test reconstruct(M, Y[:,1]) ≈ R * Y[:,1]
-@test reconstruct(M, Y) ≈ R * Y
+    T = inv(I+W'*diagm(0 => 1 ./ var(M))*W)*W'*diagm(0 => 1 ./ var(M))
+    @test transform(M, X[:,1]) ≈ T * X[:,1]
+    @test transform(M, X) ≈ T * X
 
+    R = cov(M)*W*inv(W'W)
+    @test reconstruct(M, Y[:,1]) ≈ R * Y[:,1]
+    @test reconstruct(M, Y) ≈ R * Y
 
-## PCA with non-zero mean
 
-mval = rand(5)
-M = FactorAnalysis(mval, W, Ψ)
+    ## PCA with non-zero mean
 
-@test indim(M) == 5
-@test outdim(M) == 3
-@test mean(M) == mval
-@test loadings(M) == W
-@test var(M) == Ψ
+    mval = rand(5)
+    M = FactorAnalysis(mval, W, Ψ)
 
-@test transform(M, X[:,1]) ≈ T * (X[:,1] .- mval)
-@test transform(M, X) ≈ T * (X .- mval)
+    @test indim(M) == 5
+    @test outdim(M) == 3
+    @test mean(M) == mval
+    @test loadings(M) == W
+    @test var(M) == Ψ
 
-@test reconstruct(M, Y[:,1]) ≈ R * Y[:,1] .+ mval
-@test reconstruct(M, Y) ≈ R * Y .+ mval
+    @test transform(M, X[:,1]) ≈ T * (X[:,1] .- mval)
+    @test transform(M, X) ≈ T * (X .- mval)
 
+    @test reconstruct(M, Y[:,1]) ≈ R * Y[:,1] .+ mval
+    @test reconstruct(M, Y) ≈ R * Y .+ mval
 
-## prepare training data
 
-d = 5
-n = 1000
+    ## prepare training data
 
-R = collect(qr(randn(d, d)).Q)
-@test R'R ≈ Matrix(I, 5, 5)
-rmul!(R, Diagonal(sqrt.([0.5, 0.3, 0.1, 0.05, 0.05])))
+    d = 5
+    n = 1000
 
-X = R'randn(5, n) .+ randn(5)
-mval = vec(mean(X, dims=2))
-Z = X .- mval
+    R = collect(qr(randn(d, d)).Q)
+    @test R'R ≈ Matrix(I, 5, 5)
+    rmul!(R, Diagonal(sqrt.([0.5, 0.3, 0.1, 0.05, 0.05])))
 
-# facm (default) & faem
-fa_methods = [:cm, :em]
+    X = R'randn(5, n) .+ randn(5)
+    mval = vec(mean(X, dims=2))
+    Z = X .- mval
 
-fas = FactorAnalysis[]
-for method in fa_methods
-    let M = fit(FactorAnalysis, X, method=method, maxiter=5000),
-        P = projection(M),
-        W = loadings(M)
-    push!(fas,M)
+    # facm (default) & faem
+    fa_methods = [:cm, :em]
 
-    @test indim(M) == 5
-    @test outdim(M) == 4
-    @test mean(M) == mval
-    @test P'P ≈ Matrix(I, 4, 4)
-    @test all(isapprox.(cov(M), cov(X, dims=2), atol=1e-3))
+    fas = FactorAnalysis[]
+    for method in fa_methods
+        let M = fit(FactorAnalysis, X, method=method, maxiter=5000),
+            P = projection(M),
+            W = loadings(M)
+        push!(fas,M)
 
-    M = fit(FactorAnalysis, X; mean=mval, method=method)
-    @test loadings(M) ≈ W
+        @test indim(M) == 5
+        @test outdim(M) == 4
+        @test mean(M) == mval
+        @test P'P ≈ Matrix(I, 4, 4)
+        @test all(isapprox.(cov(M), cov(X, dims=2), atol=1e-3))
 
-    M = fit(FactorAnalysis, Z; mean=0, method=method)
-    @test loadings(M) ≈ W
+        M = fit(FactorAnalysis, X; mean=mval, method=method)
+        @test loadings(M) ≈ W
 
-    M = fit(FactorAnalysis, X; maxoutdim=3, method=method)
-    P = projection(M)
+        M = fit(FactorAnalysis, Z; mean=0, method=method)
+        @test loadings(M) ≈ W
 
-    @test indim(M) == 5
-    @test outdim(M) == 3
-    @test P'P ≈ Matrix(I, 3, 3)
+        M = fit(FactorAnalysis, X; maxoutdim=3, method=method)
+        P = projection(M)
+
+        @test indim(M) == 5
+        @test outdim(M) == 3
+        @test P'P ≈ Matrix(I, 3, 3)
+        end
     end
-end
 
-# compare two algorithms
-M1, M2 = fas
-@test all(isapprox.(cov(M1), cov(M2), atol=1e-3)) # noise
-LL(m, x) = (-size(x,2)/2)*(size(x,1)*log(2π) + log(det(cov(m))) + tr(inv(cov(m))*cov(x, dims=2)))
-@test LL(M1, X) ≈ LL(M2, X) # log likelihood
-
-# test that fit works with Float32 values
-X2 = convert(Array{Float32,2}, X)
-# Float32 input
-M = fit(FactorAnalysis, X2; method=:cm, maxoutdim=3)
-M = fit(FactorAnalysis, X2; method=:em, maxoutdim=3)
-
-# views
-M = fit(FactorAnalysis, view(X2, :, 1:100), method=:cm, maxoutdim=3)
-M = fit(FactorAnalysis, view(X2, :, 1:100), method=:em, maxoutdim=3)
+    # compare two algorithms
+    M1, M2 = fas
+    @test all(isapprox.(cov(M1), cov(M2), atol=1e-3)) # noise
+    LL(m, x) = (-size(x,2)/2)*(size(x,1)*log(2π) + log(det(cov(m))) + tr(inv(cov(m))*cov(x, dims=2)))
+    @test LL(M1, X) ≈ LL(M2, X) # log likelihood
+
+    # test that fit works with Float32 values
+    X2 = convert(Array{Float32,2}, X)
+    # Float32 input
+    M = fit(FactorAnalysis, X2; method=:cm, maxoutdim=3)
+    M = fit(FactorAnalysis, X2; method=:em, maxoutdim=3)
+
+    # views
+    M = fit(FactorAnalysis, view(X2, :, 1:100), method=:cm, maxoutdim=3)
+    M = fit(FactorAnalysis, view(X2, :, 1:100), method=:em, maxoutdim=3)
+
+end