Non-convergence of Conditional Autoregressive model

Questions
9

I was thinking more of normalising it in the definition of logp:

clase ...
    def logp(self, x):
        ...
        mu_w = mu_w - mu_w.mean()

        return pm.Normal.dist(mu=mu_w, tau=priorvar).logp(x)

and with

     mu = tt.exp(b0 + mu_phi  + tt.log(offset))

However doing so seems to render rho unidentifiable.