Tracking session in gpflow

[JSP21]

With respect to the code in https://github.com/GPflow/GPflow/blob/develop/gpflow/conditionals.py
what is the proper way to tap the session of Xnew inside the: "@conditional.register(object, object, Kernel, object) @name_scope("conditional") def _conditional(Xnew, X, kern, f, *, full_cov=False, q_sqrt=None, white=False):" method. And how to actually view the value of Xnew in ndarray format inside _conditional method.

Though this issue is more related to gpflow, any help is highly appreciated!

[hughsalimbeni]

I'm not sure what you mean by 'tap' in this context. In general getting tensors out the graph is a little tricky: they need to be fetched using session.run, and there's no easy way to do this without writing a custom class / conditional. NB conditional itself is called several times, so the logic of fetching the relevant tensor needs to be at the model level, not inside the conditional itself.

The predict_all_layers_full_cov is included for the purpose of getting the inner layer values. Perhaps that is helpful as a starting point?

In the future tensorflow2 (and gpflow2 GPflow/GPflow#871) will address exactly this issue.