I’ve been using Stan to model a Dependent Dirichlet Process. When I run the model 10 times, only 2 of those runs show the log posterior converging or appear to converge. The results don’t improve with different iteration counts—I’ve tried 2,000, 4,000, 8,000, and 16,000 iterations, but the outcome remains similar each time. Could you suggest anything that might help resolve this issue?
It sounds like your posterior is degenerate, which suggests the model is either misspecified or miscoded. Can you post the code here?