Small edits

Author: theogf

src/hyperparameter/autotuning.jl

@@ -34,11 +34,13 @@ function update_hyperparameters!(
 )
     if !isnothing(gp.opt)
         f_l, f_μ₀ = hyperparameter_gradient_function(gp, X)
-        aduse = K_ADBACKEND[] == :auto ? ADBACKEND[] : K_ADBACKEND[]
-        grads = if aduse == :forward_diff
+        ad_use = K_ADBACKEND[] == :auto ? ADBACKEND[] : K_ADBACKEND[]
+        grads = if ad_use == :forward_diff
             ∇L_ρ_forward(f_l, gp, X)
-        elseif aduse == :reverse_diff
+        elseif ad_use == :reverse_diff
             ∇L_ρ_reverse(f_l, gp, X)
+        else
+            error("Uncompatible ADBackend")
         end
         grads[pr_mean(gp)] = f_μ₀()
         apply_grads_kernel_params!(gp.opt, kernel(gp), grads) # Apply gradients to the kernel parameters

src/hyperparameter/forwarddiff_rules.jl

@@ -45,12 +45,12 @@ function Z_gradient_forward(
     opt::InferenceOptimizer,
 ) where {T<:Real}
     gradient_inducing_points = similar(gp.Z.Z)
-    #preallocation
+    #pre-allocation
     Jmm, Jnm = indpoint_derivative(kernel(gp), gp.Z),
     indpoint_derivative(kernel(gp), X, gp.Z)
     for j = 1:gp.dim #Iterate over the points
         for k = 1:size(gp.Z, 2) #iterate over the dimensions
-            @views gradient_inducing_points[j, k] =
+            @views gradient_inducing_points[j, k] .=
                 f_Z(Jmm[:, :, j, k], Jnm[:, :, j, k], ∇E_μ, ∇E_Σ, i, opt)
         end
     end
@@ -71,7 +71,7 @@ function Z_gradient_forward(
     indpoint_derivative(kernel(gp), gp.Z)
     for j = 1:gp.dim #Iterate over the points
         for k = 1:size(gp.Z, 2) #iterate over the dimensions
-            @views Z_gradient[j, k] = f_Z(
+            @views Z_gradient[j, k] .= f_Z(
                 Jmm[:, :, j, k],
                 Jnm[:, :, j, k],
                 Jab[:, :, j, k],