Got you, I have few questions based on the above discussion-
- So from what I understand, logp and conditional_logp would give the same thing if used for just one variable, is that right? I want to make sure they both are the same because in my case I am trying to extract conditional logp of X_t given X_t-1. In my model, only y is observed and the rest of the variables are estimated.
- In the above model, I have obs = pm.Poisson(‘obs’, f(ar_states,z), observed=data_out, total_size=424560). If I add error = pm.Potential(pm.logp((X_t, model.rvs_to_values[X_t]))) then it will override the Poisson term, right? or the model will create a potential for Poisson and add it to the error potential?