Merge pull request #21 from xukai92/generic-gramian

Make sure gaussian_gramian is generic

src/utils.jl

@@ -2,7 +2,8 @@
 # Licensed under the MIT License. See LICENSE in the project root.
 # ------------------------------------------------------------------
 
+euclidsq(x, y) = sum((x[i] - y[i])^2 for i in eachindex(x))
+
 function gaussian_gramian(xs, ys; σ=1)
-  euclidsq(x,y) = sum((x .- y).^2)
-  [exp(-euclidsq(x,y)/2σ^2) for x in xs, y in ys]
+  [exp(-euclidsq(x, y) / 2σ^2) for x in xs, y in ys]
 end

test/basic.jl

@@ -1,18 +1,25 @@
 @testset "Basic" begin
-  for (d_nu, d_de) in [pair₁, pair₂]
-    Random.seed!(123)
-    x_nu, x_de = rand(d_nu, 100), rand(d_de, 200)
+  @testset "Gramian" begin
+    x_nu, x_de = [rand(2) for i=1:100], [rand(2) for i=1:200]
+    G = DensityRatioEstimation.gaussian_gramian(x_nu, x_de, σ=1.0)
+    @test size(G) == (length(x_nu), length(x_de))
+    @test all(G .> 0)
 
-    @testset "Gramian" begin
-      G = DensityRatioEstimation.gaussian_gramian(x_nu, x_de, σ=1.0)
-      @test size(G) == (length(x_nu), length(x_de))
-      @test all(G .> 0)
+    G = DensityRatioEstimation.gaussian_gramian(x_nu, x_nu, σ=2.0)
+    @test issymmetric(G)
+    @test all(G .> 0)
 
-      G = DensityRatioEstimation.gaussian_gramian(x_nu, x_nu, σ=2.0)
-      @test issymmetric(G)
-      @test all(G .> 0)
-    end
+    # features can be any indexable
+    x_nu = [(a=1.,b=2.),(a=3.,b=4.)]
+    x_de = [(a=1.,b=2.),(a=3.,b=4.),(a=5.,b=6.)]
+    G = DensityRatioEstimation.gaussian_gramian(x_nu, x_de)
+    @test size(G) == (2, 3)
+    @test all(G .> 0)
+  end
 
+  for (d_nu, d_de) in [pair₁, pair₂]
+    Random.seed!(123)
+    x_nu, x_de = rand(d_nu, 100), rand(d_de, 200)
     @testset "$dre -- $optlib" for (dre, optlib) in [(KMM(),   JuMPLib),
                                                      (KLIEP(), OptimLib),
                                                      (KLIEP(), ConvexLib)]