Autograd usage problem

[fatihtalu]

I want to generate the gradient value at [1,2]. I wrote the following code to do this, but it gives an error. Thanks in advance for your help.

using Knet
J(w) = w[1]^2 .+ w[2]^2;
dJ = grad(J);
w = [1,2];
println(J(w))
println(dJ(w))

Output:
5
MethodError: ^(::AutoGrad.Rec{Int64}, ::Int64) is ambiguous. Candidates:
^(x, p::Integer) in Base at intfuncs.jl:199
^(x1::AutoGrad.Rec{##1045}, x2::##1046) where {##1045<:Number, ##1046<:Number} in AutoGrad
Possible fix, define
^(::AutoGrad.Rec{##1045<:Number}, ::Integer)

[fatihtalu]

Trying with broadcast, it gives the following error:
println(dJ.(w))
Output:
BoundsError

[CarloLucibello]

This has been fixed recently in master (Ref. #48) . Try after

Pkg.checkout("AutoGrad")
Pkg.checkout("Knet")

As an alternative, you could consider evaluating in [1.0, 2.0]

[fatihtalu]

Thanks CarloLucibello.
After tried [1.0, 2.0], the problem is solved.
Output:

5.0
[2.0, 4.0]