Error while concatenating TrackedArray with normal array

[efmanu]

Hi,

In my model, I am trying to vector concatenate arrays like below:

x_rnd ~ MvNormal(x,fill(0.1,3))
 z_rnd ~ MvNormal(z,fill(0.1,3))
y = rand(3)
 a_rnd = vcat(x_rnd, y,z_rnd)

I am getting the following error while sampling by using reverse differentiation AD backend. 
ERROR: MethodError: vcat(::ReverseDiff.TrackedArray{Float64,Float64,1,Array{Float64,1},Array{Float64,1}}, ::Array{Float64,1}, ::ReverseDiff.TrackedArray{Float64,Float64,1,Array{Float64,1},Array{Float64,1}}) is ambiguous. Candidates:
  vcat(611::ReverseDiff.TrackedArray, 612::AbstractArray, x::Union{ReverseDiff.TrackedArray, ReverseDiff.TrackedReal}, xs::Union{Number, AbstractArray}...) in DistributionsAD.ReverseDiffX at C:\Users\manoos\.julia\packages\DistributionsAD\xJ5bm\src\reversediffx.jl:43
  vcat(V::AbstractArray{T,1} where T...) in Base at abstractarray.jl:1287
  vcat(x::Union{ReverseDiff.TrackedArray, ReverseDiff.TrackedReal}, xs::Union{Number, AbstractArray}...) in DistributionsAD.ReverseDiffX at C:\Users\manoos\.julia\packages\DistributionsAD\xJ5bm\src\reversediffx.jl:43
  vcat(A::Union{AbstractArray{T,2}, AbstractArray{T,1}} where T...) in Base at abstractarray.jl:1350
  vcat(609::AbstractArray, 610::AbstractArray, x::Union{ReverseDiff.TrackedArray, ReverseDiff.TrackedReal}, xs::Union{Number, AbstractArray}...) in DistributionsAD.ReverseDiffX at C:\Users\manoos\.julia\packages\DistributionsAD\xJ5bm\src\reversediffx.jl:43
  vcat(A::Union{LinearAlgebra.UniformScaling, Union{AbstractArray{T,2}, AbstractArray{T,1}} where T}...) in LinearAlgebra at D:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.3\LinearAlgebra\src\uniformscaling.jl:310
Possible fix, define
  vcat(::ReverseDiff.TrackedArray{V,D,1,VA,DA} where DA where VA where D where V, ::AbstractArray{T,1} where T, ::ReverseDiff.TrackedArray{V,D,1,VA,DA} where DA where VA where D where V, ::Vararg{AbstractArray{T,1} where T,N} where N)

The MWE for the above issue is as below:

using Turing
using DistributionsAD
using LazyArrays
using ReverseDiff
dist_f(a,b) = sqrt(sum((a .- b).^2))
function mean_y(a,b,r)
	y_m = map(dist_f,a,b)
	y_c = y_m.*r
	return y_c
end

function extract_cords(a) 
	(i,j,k) =  map(x->getindex.(a, x), 1:3)
	return vec(i), vec(j), vec(k)
end

@model test_samp(y, a, b, x, y1, z,::Type{T}=Float64) where {T} = begin
	mn = length(x)
	r_dis ~ filldist(Uniform(0.1, 1.0), 4)
	x_rnd ~ MvNormal(x,fill(0.1,mn))
	# y_rnd ~ MvNormal(vec(a_rn[2,:]),fill(0.1,mn))
    z_rnd ~ MvNormal(z,fill(0.1,mn))

    # a_rnd = vcat(vcat(x_rnd, y1), z_rnd)
    a_rnd = vcat(x_rnd, y1,z_rnd)
	mean_arnd = reshape(a_rnd, (3,mn))
	resha_mean_arnd = reshape([(mean_arnd[1,i], mean_arnd[2,i], mean_arnd[3,i]) for i in 1:size(mean_arnd, 2)], size(a))
	mean_proj_img = mean_y(resha_mean_arnd, b,  reshape(r_dis,(2,2)))
    dists = DistributionsAD.lazyarray(Normal, copy(mean_proj_img), 0.1)
    y ~ arraydist(dists)
end
r = rand(Float64,2,2);
a = fill((1.0, 2.0, 0.2), 2, 2)
b = fill((0.0, 0.1, 0.6), 2, 2)
(x,y,z) = extract_cords(a)
out = rand(size(a,1),size(a,2))
model = test_samp(out, a, b, x, y, z)
Turing.setadbackend(:reversediff)
chain = sample(model, NUTS(0.65), 100)

Regards,
Manu