Comparing lazy execution to eager execution

The main change in TensorFlow 2 is eager execution. Historically, TensorFlow 1 always used lazy execution by default. It is called lazy because operations are not run by the framework until asked specifically to do so.

Let's start with a very simple example to illustrate the difference between lazy and eager execution, summing the values of two vectors:

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([0, 0, 1])
c = tf.add(a, b)

print(c)

Note that tf.add(a, b) could be replaced by a + b since TensorFlow overloads many Python operators.

The output of the previous code depends on the TensorFlow version. With TensorFlow 1 (where lazy execution is the default mode), the output would be this:

Tensor("Add:0", shape=(3,), dtype=int32)

However, with TensorFlow 2 (where eager execution is the default mode), you would get the following output:

tf.Tensor([1 2 4], shape=(3,), dtype=int32)

In both...