In your case the logp seems to come out fine though.
I would still suggest you go with model 2 that uses Potential, since you are directly interfering with the logp component of the model. We may batch the observed in the future.
In your case the logp seems to come out fine though.
I would still suggest you go with model 2 that uses Potential, since you are directly interfering with the logp component of the model. We may batch the observed in the future.