Support for geodesic accelaration with vector function

[albertomercurio]

Hello,

I'm yet able to fit a vector model, that is model(x, p) returns an array of dimensions (2, N), where N is the dimension of the x data points. I'm yet using forward differentiation ad a numerical computed Jacobian

model_jac(x, p) = ForwardDiff.jacobian(z->model(x, z), p)

and it fits very well.

Now I want to use geodesic acceleration to see if I can increase the accuracy. Taking the example in the Readme file, in which they calculate the hessian with hessian(z->model(in_data[i], z),p), I needed first to define the hessian of a vector function, which, following this issue, can be defined as

function vector_hessian(f, x)
    n = length(x)
    out = ForwardDiff.jacobian(x -> ForwardDiff.jacobian(f, x), x)
    return reshape(out, n, n, 2) # here 2 is the dimension that my function returns since I calculate the hessian(z->model(in_data[i], z),p)
end

However, if I declare my Avv! function with this 'vectorized' hessian, it gives me the error

MethodError: no method matching *(::Matrix{Float64}, ::Array{Float64, 3})

since my hessian is not a Matrix anymore.

Is there any way to make this simple extension work?