Model validation

What would be the the preferred Pyro way for model validation? My typical statistical way would be to split the data into training and testing, train on training set and test (validate) on the testing set.

In fact, my question is similar to:

Predictive seems to be the right way, but then I would have the same question as nitishp25 regarding how pass the testing set?

I think I can create a new pyro.condition function to pass the testing dataset. But it this case, I will face another challenge since I define the shape of the tensors in Model and Guide using training set shape.

Another way (I think less elegant) is just to use estimated parameters and sample using model equations.


@Vlad See my answer on that topic. :slight_smile: