Removing unnecessary functions and uncommented all cases

Author: theogf

test/test_AD.jl

@@ -1,7 +1,8 @@
 using KernelFunctions
-using KernelFunctions: kappa
+using KernelFunctions: kappa, ColVecs, RowVecs
 using Flux: params
 import Zygote, ForwardDiff, ReverseDiff
+using Zygote: pullback
 using Test, LinearAlgebra, Random
 using FiniteDifferences
 
@@ -26,10 +27,10 @@ kernels = [
     SqExponentialKernel(),
     ExponentialKernel(),
     MaternKernel(ν = ν),
-    # transform(SqExponentialKernel(), l),
-    # transform(SqExponentialKernel(), vl),
-    # ExponentiatedKernel() + LinearKernel(),
-    # 2.0 * PolynomialKernel() * Matern32Kernel(),
+    transform(SqExponentialKernel(), l),
+    transform(SqExponentialKernel(), vl),
+    ExponentiatedKernel() + LinearKernel(),
+    2.0 * PolynomialKernel() * Matern32Kernel(),
 ]
 
 ds = log.([eps(), rand(rng)])

test/utils_AD.jl

@@ -1,4 +1,3 @@
-allapprox(x, y, tol = 1e-8) = all(isapprox.(x, y, atol = tol))
 FDM = central_fdm(5, 1)
 
 function gradient(::Val{:Zygote}, f::Function, args)
@@ -20,51 +19,3 @@ end
 function gradient(::Val{:FiniteDiff}, f::Function, args)
     first(FiniteDifferences.grad(FDM, f, args))
 end
-
-
-
-function transform_AD(::Val{:Zygote}, t::Transform, A)
-    ps = KernelFunctions.params(t)
-    @test allapprox(
-        first(Zygote.gradient(p -> transform_with_duplicate(p, t, A), ps)),
-        first(FiniteDifferences.grad(
-            FDM,
-            p -> transform_with_duplicate(p, t, A),
-            ps,
-        )),
-    )
-    @test allapprox(
-        first(Zygote.gradient(X -> sum(transform(t, X, 2)), A)),
-        first(FiniteDifferences.grad(FDM, X -> sum(transform(t, X, 2)), A)),
-    )
-end
-
-function transform_AD(::Val{:ForwardDiff}, t::Transform, A)
-    ps = KernelFunctions.params(t)
-    if t isa ScaleTransform
-        @test allapprox(
-            first(ForwardDiff.gradient(
-                p -> transform_with_duplicate(first(p), t, A),
-                [ps],
-            )),
-            first(FiniteDifferences.grad(
-                FDM,
-                p -> transform_with_duplicate(p, t, A),
-                ps,
-            )),
-        )
-    else
-        @test allapprox(
-            ForwardDiff.gradient(p -> transform_with_duplicate(p, t, A), ps),
-            first(FiniteDifferences.grad(
-                FDM,
-                p -> transform_with_duplicate(p, t, A),
-                ps,
-            )),
-        )
-    end
-    @test allapprox(
-        ForwardDiff.gradient(X -> sum(transform(t, X, 2)), A),
-        first(FiniteDifferences.grad(FDM, X -> sum(transform(t, X, 2)), A)),
-    )
-end