Update to AutoDiffOperators v0.2

Project.toml

@@ -26,7 +26,7 @@ Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 ValueShapes = "136a8f8c-c49b-4edb-8b98-f3d64d48be8f"
 
 [compat]
-AutoDiffOperators = "0.1.3"
+AutoDiffOperators = "0.1.3, 0.2"
 ChainRulesCore = "0.9.44, 0.10, 1"
 Distributed = "1"
 Distributions = "0.25"

docs/src/advanced_tutorial_lit.jl

@@ -36,7 +36,7 @@ using FFTW
 import ForwardDiff, Zygote
 using AutoDiffOperators
 
-context = MGVIContext(ADModule(:Zygote))
+context = MGVIContext(ADSelector(Zygote))
 
 #-
 Random.seed!(84612);

docs/src/tutorial_lit.jl

@@ -19,7 +19,7 @@ import Zygote
 #
 # We assume errors are normally distributed with unknown covariance, which has to be learned as well.
 
-context = MGVIContext(ADModule(:Zygote))
+context = MGVIContext(ADSelector(Zygote))
 
 const _x1_grid = [Float64(i)/10 for i in 1:25]
 const _x2_grid = [Float64(i)/10 + 0.1 for i in 1:15]

src/mgvi_impl.jl

@@ -42,7 +42,7 @@ normal distribution of the same dimensionality as `init_param_point`.
 using Random, Distributions, MGVI
 import Zygote
 
-context = MGVIContext(ADModule(:Zygote))
+context = MGVIContext(ADSelector(Zygote))
 
 model(x::AbstractVector) = Normal(x[1], 0.2)
 true_param = [2.0]

test/test_jacobians.jl

@@ -21,9 +21,9 @@ Test.@testset "test_jacobians_consistency" begin
         x = rand(6); l = rand(4); r = rand(6)
         J_ref = ForwardDiff.jacobian(f, x)
 
-        _, J1 = @inferred with_jacobian(f, x, Matrix, ADModule(:ForwardDiff))
-        _, J2 = @inferred with_jacobian(f, x, LinearMap, ADModule(:Zygote))
-        _, J3 = @inferred with_jacobian(f, x, LinearMap, ADModule(:ForwardDiff))
+        _, J1 = @inferred with_jacobian(f, x, Matrix, ADSelector(ForwardDiff))
+        _, J2 = @inferred with_jacobian(f, x, LinearMap, ADSelector(Zygote))
+        _, J3 = @inferred with_jacobian(f, x, LinearMap, ADSelector(ForwardDiff))
 
         for J in (J1, J2, J3)
             @test @inferred(Matrix(J)) ≈ J_ref
@@ -38,9 +38,9 @@ Test.@testset "test_jacobians_consistency" begin
     _flat_model = MGVI.flat_params ∘ ModelPolyfit.model
     true_params = ModelPolyfit.true_params
 
-    _, full_jac = @inferred with_jacobian(_flat_model, true_params, Matrix, ADModule(:ForwardDiff))
-    _, fwdder_jac = @inferred with_jacobian(_flat_model, true_params, LinearMap, ADModule(:ForwardDiff))
-    _, fwdrevad_jac = @inferred with_jacobian(_flat_model, true_params, LinearMap, ADModule(:Zygote))
+    _, full_jac = @inferred with_jacobian(_flat_model, true_params, Matrix, ADSelector(ForwardDiff))
+    _, fwdder_jac = @inferred with_jacobian(_flat_model, true_params, LinearMap, ADSelector(ForwardDiff))
+    _, fwdrevad_jac = @inferred with_jacobian(_flat_model, true_params, LinearMap, ADSelector(Zygote))
 
     for i in 1:min(size(full_jac)...)
         vec = rand(size(full_jac, 2))

test/test_mgvi_impl.jl

@@ -13,7 +13,7 @@ if !isdefined(Main, :ModelPolyfit)
 end
 
 Test.@testset "test_mgvi_optimize_step" begin
-    context = MGVIContext(ADModule(:Zygote))
+    context = MGVIContext(ADSelector(Zygote))
 
     model = ModelPolyfit.model
     true_params = ModelPolyfit.true_params

test/test_samplers.jl

@@ -14,7 +14,7 @@ if :ModelPolyfit ∉ names(Main)
 end
 
 Test.@testset "test_cmp_residual_samplers" begin
-    context = MGVIContext(ADModule(:Zygote))
+    context = MGVIContext(ADSelector(Zygote))
 
     model = ModelPolyfit.model
     true_params = ModelPolyfit.true_params