sample_as_parent in variational_gamma

[hyanwong]

One of the tests for the inside_outside method is test_sample_as_parent_fails. Can we actually date samples that have descendant undated nodes using the current variational_gamma implementation, or should we be failing in this case too, @nspope ?

[nspope]

The machinery is (mostly) there but it's not working yet. Do we need to add an exception so that it errors out appropriately?

[hyanwong]

Right, I guess it would be helpful to error out, but it's not a high priority. Does the same go for non-contemporary samples?

[nspope]

Yup, same deal -- EP machinery in place, but rescaling routine doesn't do the right thing yet (although it's conceptually very straightforward to incorporate). Not going to worry about getting all that sorted out until we have a solid application.