How to apply multiple tensorflow functions to the same tensor in parallel?

Question

Let t be a Tensorflow tensor of shape (m, n) . Let F be a list of p functions f_1, f_2, ... where each function of these takes a tensor and outputs a tensor of the same shape.

I need to compute the tensor T of shape (p, m, n) such that T[i]=f_i(t) .

Notice that this is the inverse setup of the map function where we apply 1 function to a collection of element. Here, we need to apply a collection of functions to 1 element. Additionally, we need to do that in parallel .

The naive approach (non-parallel) would be stacking results of these computations generated via a for loop (or a list comprehension).

T = tf.stack([
  f(t) for f in F
])

Tensorflow version: 2.0.0a. Mode: eager

Answer 1

假设r1.13不急于执行tf.parallel_stack应该做的伎俩。