fix most tests

Manifest.toml

@@ -309,7 +309,7 @@ version = "0.8.1"
 
 [[Zygote]]
 deps = ["DiffRules", "ForwardDiff", "IRTools", "InteractiveUtils", "LinearAlgebra", "MacroTools", "NNlib", "NaNMath", "Random", "Requires", "SpecialFunctions"]
-git-tree-sha1 = "7fcb55117550e1c195a646947135cc9aac1e2afc"
+git-tree-sha1 = "db27148be2365d2fe507f49ada875050b08d8187"
 repo-rev = "master"
 repo-url = "https://github.com/FluxML/Zygote.jl.git"
 uuid = "e88e6eb3-aa80-5325-afca-941959d7151f"

test/cuda/cuda.jl

@@ -1,4 +1,4 @@
-using Flux, Flux.Tracker, CuArrays, Test
+using Flux, CuArrays, Test
 using Flux: gpu
 
 @info "Testing GPU Support"

test/cuda/cudnn.jl

@@ -1,5 +1,4 @@
-using Flux, Flux.Tracker, CuArrays, Test
+using Flux, CuArrays, Test
-using Flux.Tracker: TrackedArray, data
 
 @testset "CUDNN BatchNorm" begin
     @testset "4D Input" begin

test/layers/normalisation.jl

@@ -1,5 +1,4 @@
 using Flux: testmode!
-using Flux.Tracker: data
 
 @testset "Dropout" begin
   x = [1.,2.,3.]
@@ -29,8 +28,8 @@ using Flux.Tracker: data
 end
 
 @testset "BatchNorm" begin
-  let m = BatchNorm(2), x = param([1 3 5;
+  let m = BatchNorm(2), x = [1 3 5;
-                                   2 4 6])
+                             2 4 6]
 
     @test m.β.data == [0, 0]  # initβ(2)
     @test m.γ.data == [1, 1]  # initγ(2)
@@ -111,7 +110,7 @@ end
   expand_inst = (x, as) -> reshape(repeat(x, outer=[1, as[length(as)]]), as...)
   # begin tests
   let m = InstanceNorm(2), sizes = (3, 2, 2),
-      x = param(reshape(collect(1:prod(sizes)), sizes))
+      x = reshape(collect(1:prod(sizes)), sizes)
 
       @test m.β.data == [0, 0]  # initβ(2)
       @test m.γ.data == [1, 1]  # initγ(2)
@@ -157,7 +156,7 @@ end
   end
   # with activation function
   let m = InstanceNorm(2, sigmoid), sizes = (3, 2, 2),
-      x = param(reshape(collect(1:prod(sizes)), sizes))
+      x = reshape(collect(1:prod(sizes)), sizes)
 
     affine_shape = collect(sizes)
     affine_shape[1] = 1
@@ -173,7 +172,7 @@ end
   end
 
   let m = InstanceNorm(2), sizes = (2, 4, 1, 2, 3),
-      x = param(reshape(collect(1:prod(sizes)), sizes))
+      x = reshape(collect(1:prod(sizes)), sizes)
     y = reshape(permutedims(x, [3, 1, 2, 4, 5]), :, 2, 3)
     y = reshape(m(y), sizes...)
     @test m(x) == y
@@ -181,7 +180,7 @@ end
 
   # check that μ, σ², and the output are the correct size for higher rank tensors
   let m = InstanceNorm(2), sizes = (5, 5, 3, 4, 2, 6),
-      x = param(reshape(collect(1:prod(sizes)), sizes))
+      x = reshape(collect(1:prod(sizes)), sizes)
     y = m(x)
     @test size(m.μ) == (sizes[end - 1], )
     @test size(m.σ²) == (sizes[end - 1], )
@@ -190,7 +189,7 @@ end
 
   # show that instance norm is equal to batch norm when channel and batch dims are squashed
   let m_inorm = InstanceNorm(2), m_bnorm = BatchNorm(12), sizes = (5, 5, 3, 4, 2, 6),
-      x = param(reshape(collect(1:prod(sizes)), sizes))
+      x = reshape(collect(1:prod(sizes)), sizes)
     @test m_inorm(x) == reshape(m_bnorm(reshape(x, (sizes[1:end - 2]..., :, 1))), sizes)
   end
 

test/layers/stateless.jl

@@ -1,6 +1,7 @@
 using Test
 using Flux: onehotbatch, mse, crossentropy, logitcrossentropy,
             σ, binarycrossentropy, logitbinarycrossentropy
+using Zygote
 
 const ϵ = 1e-7
 
@@ -55,9 +56,9 @@ const ϵ = 1e-7
       y = rand(T, 2)
       ŷ = rand(T, 2)
       for f in (mse, crossentropy, logitcrossentropy)
-        fwd, back = Flux.Tracker.forward(mse, ŷ, y)
+        fwd, back = Zygote.forward(mse, ŷ, y)
-        @test typeof(fwd) == Flux.Tracker.TrackedReal{T}
+        @test fwd isa T
-        @test eltype(back(one(T))[1]) == Flux.Tracker.TrackedReal{T}
+        @test eltype(back(one(T))[1]) == T
       end
     end
   end

test/optimise.jl

@@ -1,13 +1,12 @@
 using Flux.Optimise
 using Flux.Optimise: runall
-using Flux.Tracker
 using Test
 @testset "Optimise" begin
   w = randn(10, 10)
   @testset for opt in [ADAMW(), ADAGrad(0.1), AdaMax(), ADADelta(0.9), AMSGrad(),
                        NADAM(), Descent(0.1), ADAM(), Nesterov(), RMSProp(),
                        Momentum()]
-    w′ = param(randn(10, 10))
+    w′ = randn(10, 10)
     loss(x) = Flux.mse(w*x, w′*x)
     for t = 1: 10^5
       θ = Params([w′])
@@ -21,7 +20,7 @@ end
 @testset "Optimiser" begin
   w = randn(10, 10)
   @testset for Opt in [InvDecay, WeightDecay, ExpDecay]
-    w′ = param(randn(10, 10))
+    w′ = randn(10, 10)
     loss(x) = Flux.mse(w*x, w′*x)
     opt = Optimiser(Opt(), ADAM(0.001))
     for t = 1:10^5
@@ -36,7 +35,7 @@ end
 
 @testset "Training Loop" begin
   i = 0
-  l = param(1)
+  l = 1
 
   Flux.train!(() -> (sleep(0.1); i += 1; l),
               (),

test/tracker.jl

@@ -1,5 +1,5 @@
 using Flux, Test
-using Tracker: gradcheck
+using Zygote: gradcheck
 
 gradtest(f, xs::AbstractArray...) = gradcheck((xs...) -> sum(sin.(f(xs...))), xs...)
 gradtest(f, dims...) = gradtest(f, rand.(Float64, dims)...)

test/utils.jl

@@ -87,12 +87,11 @@ end
 @testset "Precision" begin
   m = Chain(Dense(10, 5, relu), Dense(5, 2))
   x = rand(10)
-  @test eltype(m[1].W.data) == Float32
+  @test eltype(m[1].W) == Float32
-  @test eltype(m(x).data) == Float32
+  @test eltype(m(x)) == Float32
-  @test eltype(f64(m)(x).data) == Float64
+  @test eltype(f64(m)(x)) == Float64
-  @test eltype(f64(m)[1].W.data) == Float64
+  @test eltype(f64(m)[1].W) == Float64
-  @test eltype(f32(f64(m))[1].W.data) == Float32
+  @test eltype(f32(f64(m))[1].W) == Float32
-  @test Tracker.isleaf(f32(f64(m))[1].W)
 end
 
 @testset "Stacking" begin