TensorFlow 2.0 should be officially released this spring. I’ve started looking at TensorFlow 2.0 to some degree. There will be many breaking changes from this release. I’ve started to evaluate the impact that this will have on my Deep Learning Course at WUSTL. I mostly make use of TensorFlow through Keras, so the number of breaking changes has mainly been limited to Python import directives. For example, “import keras” becomes “import tensorflow.keras”.
Internally, the biggest change that has impact me is the eager execution. This affects the way that you make use of variables and execute a tensorflow graph. Also, Keras is now the primary interface into TensorFlow and no longer needs to be included separatly (Keras is now included with TensorFlow).
Most examples on the net that make use of Tensorflow prior to 2.0 are now broken. Tutorials for TensorFlow 2.0 are somewhat scarce, so here are some of my favorites: