In this chapter, all the major changes that were introduced in TensorFlow 2.0 have been presented, including the standardization of the framework on the Keras API specification, the way models are defined using Keras, and how to train them using a custom training loop. We even looked at graph acceleration, which was introduced by AutoGraph, and tf.function.
AutoGraph, in particular, still requires us to know how the TensorFlow graph architecture works since the Python function that's defined and used in eager mode needs to be re-engineered if there is the need to graph-accelerate them.
The new API is more modular, object-oriented, and standardized; these groundbreaking changes have been made to make the usage of the framework easier and more natural, although the subtleties from the graph architecture are still present and always will be.
Those of you who have years...