Multi target values regression with TFP

[Soroosh Shalileh]

Hi Team, 

First of all, thank you very much for the fascinating work.

Let us go to my question: 

you have explained various possibilities for probabilistic regression with TFP in the following link:

https://www.tensorflow.org/probability/examples/Probabilistic_Layers_Regression

What I am missing badly, and made me puzzled, is how should I use LambdaDistribution and the slicing techniques (e.g., t[...,:1] and t[..., 1:], in Aleatoric modeling) you have used to model mean and standard deviations? 

Similarly, how should I modify the code for Epistemic modeling?

My Best, 

-S