Automatic imputation for posterior predictive check?

Hi there,

I understand that PyMC3 does an automatic imputation of missing data. I guess this cannot be an option that you can switch off because the code would crash because of these missing values and the alternative is to mask them.

I wonder whether the automatic imputation is also done for the posterior predictive check (sample_posterior_predictive and fast_sample_posterior_predictive)?

The reason for the question is that I’ve been considering adding some deterministic variables to the trace a posteriori (post-inference) to do the posterior predictive check on them. This would prevent me from adding them to the model, since they are not useful for the inference solution and since they take a lot of memory.


Hi Vianey,
Do you have a use-case in mind that you can share? It would be easier to answer – and more concrete.
Also, note that you can disable missing data imputation if you need to.
Hope this helps :vulcan_salute: