Princeton Baretto

Results 2 comments of Princeton Baretto

![Capture](https://user-images.githubusercontent.com/43331485/82120665-7665c300-97a5-11ea-88f8-70cee9a959ce.PNG) Facing the same issue with tensorflow version 2.x

Got the solution : Just replace zero_state with get_inital_state, because the function get_initial_state returns an AttentionWrapperState tuple containing zeroed out tensors same as zero_state