Fix to_vec with Integers

Project.toml

@@ -1,6 +1,6 @@
 name = "FiniteDifferences"
 uuid = "26cc04aa-876d-5657-8c51-4c34ba976000"
-version = "0.12.18"
+version = "0.12.19"
 
 [deps]
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"

src/grad.jl

@@ -85,4 +85,4 @@ end
 
 Compute the gradient of `f` for any `xs` for which [`to_vec`](@ref) is defined.
 """
-grad(fdm, f, xs...) = j′vp(fdm, f, 1, xs...)  # `j′vp` with seed of 1
+grad(fdm, f, xs...) = j′vp(fdm, f, one(f(xs...)), xs...)  # `j′vp` with seed of 1

src/to_vec.jl

@@ -18,6 +18,11 @@ function to_vec(z::Complex)
     return [real(z), imag(z)], Complex_from_vec
 end
 
+function to_vec(x::Integer)
+    Integer_from_vec(_) = x
+    return Bool[], Integer_from_vec
+end
+
 # Base case -- if x is already a Vector{<:Real} there's no conversion necessary.
 to_vec(x::Vector{<:Real}) = (x, identity)
 

test/to_vec.jl

@@ -68,6 +68,14 @@ function test_to_vec(x::T; check_inferred=true) where {T}
 end
 
 @testset "to_vec" begin
+
+    # Integers are non-differentiable. to_vec should only preserve the differentiable bits
+    # of a type so that they can be appropriately perturbed.
+    @testset "Int" begin
+        @test isempty(to_vec(4)[1])
+        test_to_vec(5)
+    end
+
     @testset "$T" for T in (Float32, ComplexF32, Float64, ComplexF64)
         if T == Float64
             test_to_vec(1.0)