Hi, in formulating my objective function for optimization, in one of the steps, the parameters will enter a Poisson regression as the weight, which I implemented through GLM.jl package. However, when generating the gradient from the ReverseDiff.jl package, it has the following error

ERROR: StackOverflowError:

Stacktrace:

[1] GLM.GlmResp(y::Vector{Float64}, d::Poisson{Float64}, l::LogLink, off::Vector{Float64}, wts::ReverseDiff.TrackedArray{Float64, Float64, 1, Vector{Float64}, Vector{Float64}})

@ GLM ~/.julia/packages/GLM/6ycOV/src/glmfit.jl:67

[2] GLM.GlmResp(y::Vector{Float64}, d::Poisson{Float64}, l::LogLink, off::Vector{Float64}, wts::ReverseDiff.TrackedArray{Float64, Float64, 1, Vector{Float64}, Vector{Float64}})

@ GLM ~/.julia/packages/GLM/6ycOV/src/glmfit.jl:69

The MWE is attached here:

```
using DataFrames, GLM, Optim, ReverseDiff, Random, Distributions
Random.seed!(12345)
count_data = DataFrame(Y = rand(Poisson(5), 1000), X1 = rand(1:10, 1000), X2 = rand(LogNormal(3,2), 1000), X3 = randn(1000), X4 = rand(Normal(0,2), 1000), w = rand(1000))
function coef_collect(data, para)
data[:,:w_para] = data[:,:w] .*para[3]
count_coef = collect(coef(glm(@formula(Y~X1+X2+X3), data, Poisson(); wts=data[:,:w_para])))
end
function predict_data(data, para)
pred = Array(data[:,1:4])*coef_collect(data, para)
end
function optimize_func(para, y, data)
mse = mean((y.-predict_data(data, para)*para[1]+data[:,:X4].*para[2]).^2)
end
para_true = [3.0, 4.0, 1.5]
y_true = predict_data(count_data)*para_true[1]+count_data[:,:X4]*para_true[2] + randn(1000)
para_0 = [5.0,6.0, 1.3]
f_tape = ReverseDiff.GradientTape(para->optimize_func(para, y_true, count_data), para_0)
```

Thanks for your help!