Tensorflow: Tensor to numpy array conversion without running any session

Question

I have created an OP in tensorflow where for some processing I need my data to be converted from tensor object to numpy array. I know we can use tf.eval() or sess.run to evaluate any tensor object. What I really want to know is, Is there any way to convert tensor to array without any session running, so in turn we avoid use of .eval() or .run().

Any help is highly appreciated!

def tensor_to_array(tensor1):
    '''Convert tensor object to numpy array'''
    array1 = SESS.run(tensor1) **#====== need to bypass this line**
    return array1.astype("uint8")

def array_to_tensor(array):
    '''Convert numpy array to tensor object'''
    tensor_data = tf.convert_to_tensor(array, dtype=tf.float32)
    return tensor_data

Unlikely in the general case, since you don't know the answers without running the pipeline — Mad Physicist
– Mad Physicist, Commented Sep 7, 2018 at 5:19
@MadPhysicist I don't think you are making a correct statement. — R.yan
– R.yan, Commented Sep 7, 2018 at 6:10

R.yan · Accepted Answer · 2018-09-07 07:35:20Z

6

Updated

# must under eagar mode
def tensor_to_array(tensor1):
    return tensor1.numpy()

example

>>> import tensorflow as tf
>>> tf.enable_eager_execution()
>>> def tensor_to_array(tensor1):
...     return tensor1.numpy()
...
>>> x = tf.constant([1,2,3,4])
>>> tensor_to_array(x)
array([1, 2, 3, 4], dtype=int32)

I believe you can do it without tf.eval() or sess.run by using tf.enable_eager_execution()

example

import tensorflow as tf
import numpy as np
tf.enable_eager_execution()
x = np.array([1,2,3,4])
c = tf.constant([4,3,2,1])
c+x
<tf.Tensor: id=5, shape=(4,), dtype=int32, numpy=array([5, 5, 5, 5], dtype=int32)>

For more details about tensorflow eager mode, checkout here:Tensorflow eager

If without tf.enable_eager_execution():

import tensorflow as tf
import numpy as np
c = tf.constant([4,3,2,1])
x = np.array([1,2,3,4])
c+x
<tf.Tensor 'add:0' shape=(4,) dtype=int32>

edited Sep 7, 2018 at 7:35

answered Sep 7, 2018 at 6:09

R.yan

2,4121 gold badge18 silver badges36 bronze badges

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

15 Comments

KapB Over a year ago

Hi...eager_execution looks interesing, but I don't think it serves my purpose. Please have a look at below snippet.

R.yan Over a year ago

Can you provide the details of the tensor object that you would like to parse, please?

KapB Over a year ago

def tensor_to_array(tensor1): '''Convert tensor object to numpy array''' array1 = SESS.run(tensor1) #== need to replace this line === return array1.astype("uint8") def array_to_tensor(array): '''Convert numpy array to tensor object''' tensor_data = tf.convert_to_tensor(array, dtype=tf.float32) return tensor_data

KapB Over a year ago

Thanks @R.yan ...Also refer this link for further information. towardsdatascience.com/eager-execution-tensorflow-8042128ca7be

KapB Over a year ago

I'm still facing an issue. If I run all the commands from anaconda prompt, I'm getting the desired output. But when I try to run it as script in spyder, it throws an error of "AttributeError: 'Tensor' object has no attribute 'numpy'". Do you have any idea why so?? I have made sure that spyder runs from same environemnt but the problem still persists. @R.yan

|

Collectives™ on Stack Overflow

Tensorflow: Tensor to numpy array conversion without running any session

1 Answer 1

15 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

15 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related