TensorFlow2.0 会减少不必要的计算吗？

Question

Will TensorFlow 2.0 be able to trim unnecessary computations?

For example,

a = tf.constant(1.0)
for i in range(10):
  c = a + 1
  a = a + 1
print(a)

In TF1.x, c=a+1 won't be computed. Will TF2.0 automatically detect c=a+1 is not used so it doesn't do the computation?

Answer 1

It won't be computed. The preprocessor will detect it is never used and the program won't take into account. You can (not recommended) suppress warnings about unused stuff, instead, you should take care of the unused things.