Using Numpy Array Conditional Operation Mask with Tensorflow

Question

I am new to Tensoflow and I try to implement it in a custom RNN project to learn more about how Neural Networks work.

My problem is simple but I don't seem to find any satisfying answer.

I am used to Numpy and operations on arrays using condition masks but I don't find a way to transform this with Tensors

def ELu(in_array):
    in_array[in_array<= 0] = math.e ** in_array[in_array<= 0] - 1
    return in_array

>>>print(ELu(np.array([1.0,0.0,-1.0])))

Gives me

[ 1.          0.         -0.63212056]

And I would like to edit that function to be able to give me a similar Tensor if I do something like this

>>>print(ELu(tf.convert_to_tensor([1.0,0.0,-1.0])))

Which should give me

<tf.Tensor: shape=(3,), dtype=float32, numpy=array([ 1.,  0., -0.63212056], dtype=float32)>

but accessing a tensor using a similar way in_array[in_array<= 0] doesn't work

Andrey · Accepted Answer · 2021-01-17 17:54:20Z

2

Use tensor_scatter_nd_update():

import math
def ELu(in_array):
  mask = in_array <= 0
  inds = tf.where(mask)
  updates = tf.boolean_mask(in_array, mask)
  updates = math.e ** updates - 1.
  res = tf.tensor_scatter_nd_update(in_array, inds, updates)
  return res

answered Jan 17, 2021 at 17:54

Andrey

6,6393 gold badges24 silver badges51 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Julien Alardot Over a year ago

Thank you, that works perfectly. I'll start using those functions more often!

Collectives™ on Stack Overflow

Using Numpy Array Conditional Operation Mask with Tensorflow

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related