Turing NUTS chains getting stuck at the parameter bounds

@Christopher_Fisher I think I know what caused the Loglik surface to be wobbly in your code: parms = p modifies also p during the mapping. If you do parms = copy(p) instead, the surfaces look more like they should. Beta still shows a flat surface, with the max LL at the true value of 0.15, but values > 0.15 also have a high LL, so I can see why that is difficult for the sampler.
beta
eta
phi
psi

2 Likes