Hi EvoArt:
Thank you very much for looking into it. I did have the code to categorized CMetastized:
HSAUR[!,:CMetastized]=levelcode.(HSAUR[!,:Metastized]).-1;
I forgot to post that I used CategoricalArrays.jl . But your method works easier and faster.
And thanks again for your explanation. I agree with you, since the exposure has many 0, that sampler would be struggled with… the formula is: μ = exposureλ = exposureλ0 exp.(β . x) that’s why posterior is Poisson. shift by 0.1 should not affect the result. I don’t know how PyMC3 samples such a scenario and R also has a package, I will see if I can find what posterior other people used from paper. I tried to run the PyMC3 code, but it has many package dependency issues.
Best Regards,
Ryan