Leave-one-out cross validation returns nan

I have set up a model for Poisson regression (with Normal priors, similar to the example in the documentation, but without using patsy). The trace plots after sampling look reasonable. However, when I run the loo() function to do cross-validation, the returned values are all nan. In addition, I get the UserWarning: “Estimated shape parameter of Pareto distribution is greater than 0.7 for one or more samples.” followed by the recommendation to “consider using a more robust model”. It is unclear to me how I should change the model to make it more “robust” and get the leave-one-out cross validation to work. Any suggestions?

Hi @fkloos, could you please provide the data and model (ideally in a self-contain ipynb)? It is a bit difficult to say about the NaN problem without more information.