There are no relevant documents available to address your query about sampling options with many observations and a HurdleGamma likelihood. However, based on general knowledge, when dealing with divergences in sampling, especially with complex likelihoods like HurdleGamma, you might consider the following approaches:
-
Reparameterization: Sometimes, reparameterizing the model can help reduce divergences. This involves changing the parameterization of your model to make it more stable numerically.
-
Tuning: Increase the number of tuning steps. This allows the sampler more time to adapt to the posterior distribution.
-
Step Size Adjustment: Manually adjusting the step size can sometimes help. A smaller step size might reduce divergences at the cost of longer sampling times.
-
Prior Specification: Ensure that your priors are not too tight or too vague, as this can affect the sampling process. Even though you mentioned not having a prior on the psi parameter, consider if a weakly informative prior might help.
-
Alternative Samplers: If you haven’t already, try using different samplers or algorithms that might be more robust to the specific challenges of your model.
-
Model Simplification: Simplifying the model, if possible, can sometimes help identify the source of the problem.
If these suggestions do not resolve the issue, it might be helpful to consult with others who have experience with similar models or to provide more details about the model structure and data.
For more information, visit https://www.pymc.io/
This answer was generated by the PyMC Assistant bot in response to your command. Please let us know if it was helpful!