I am trying to fit a model of the form:
for(i in 1:N)
X[i] ~ multinomial(theta);
theta ~ dirichlet(exp(alpha));
alpha ~ normal(0,1);
where alpha is a univariate parameter. Unsurprisingly, this model – being non-centred – suffers from divergent iterations. These do not go away if I increase adapt_delta nor if I decrease the step size. I think I need a non-centred parameterisation here.
I’ve had a look at can’t seem to find one that people have used in the past for Dirichlet distributions. Does anyone have an idea how to ‘non-centre’ the above?
[edit: escaped code]