It currently has the data normally distributed around the ODE solution. However, given that this is an ecological model, one would never actually observe negative densities or counts of animals in the wild. (The existence of atto-foxes is problematic enough without positing the existence of theoretical anti-foxes...) Would it make sense to make the data and observation distributions Poisson counts instead? I just tried it out and the model still converges easily.
A question/suggestion on this example: https://github.com/TuringLang/Turing.jl/edit/master/docs/_tutorials/10_bayesian-differential-equations.md
It currently has the data normally distributed around the ODE solution. However, given that this is an ecological model, one would never actually observe negative densities or counts of animals in the wild. (The existence of atto-foxes is problematic enough without positing the existence of theoretical anti-foxes...) Would it make sense to make the data and observation distributions Poisson counts instead? I just tried it out and the model still converges easily.