We're starting down the road to TensorFlow 2.0. Now's the time for contributors and interested parties to pay attention to the developer mailing list, for feedback on public design reviews: more info belowhttps://groups.google.com/a/tensorflow.org/d/msgid/announce/CADtzJKOYMjX-GiF1tnmJO3ORBoE9KH%2BOwrgObz8UfzvLOec2XQ%40mail.gmail.com?utm_medium=email&utm_source=footer …
I _liked_ the compute graph API. It was elegant and easy to understand. The only problem was that the Python APIs sometimes tried to pretend that TF was imperative, when it was not. This caused confusion.