What's the purpose of the "predictions" argument in the pymc.sample_posterior_predictive function?


Is there any reference that elaborates on the purpose of the predictions argument in the sample_posterior_predictive function? By default it is set to False and I thought that I could use the values in trace.posterior_predictive to get my predictions. However, after diving into the Forecasting with Structural AR Timeseries notebook, I found out the author uses predictions=True and trace.predictions, and this produces wider intervals in the prediction plots.

Any guidance is appreciated.

This question has been answered here. For convenience, I’m sharing the same screenshot with the explanation:

1 Like