I am looking for a convenient way of calculation data logp after obtaining VI posterior. I’ve noticed that inferece.objective.op.datalogp_norm looks like something I may use. However, when I try to run .eval() on this object, theano fails with:
MissingInputError: Input 1 of the graph (indices start from 0), used to compute Elemwise{mul,no_inplace}(InplaceDimShuffle{x,0}.0, MeanFieldGroup_symbolic_initial_tensor), was not provided and not given a value. Use the Theano flag exception_verbosity=‘high’, for more information on this error.
I am not 100% sure I understand what is happening here, is it expected behaviour?