why random gives different result with global and operation level seed pre-set

[chenalb0075]

Hi, i am just started to learn tensorflow. From the document i know that when global and operation level seed were set, i supposed to generate the same tensor value. But when i run following code on colab, it always gave different value.

code:
import tensorflow as tf
tf.random.set_seed=1234
print(tf.random.normal([2,2],seed=18))
print(tf.random.normal([2,2],seed=18))

Can somebody tell me what's wrong here?

[emrecicekyurt]

Hi,

If you check the documentation (https://www.tensorflow.org/api_docs/python/tf/random/set_seed) of tf.random.set_seed, you will see that the second print method uses a different operation seed. If you want to get the same tensor, you need to assign the tf.random.set_seed again with the same value "1234" as below

tf.random.set_seed(1234)
print(tf.random.uniform([1])) # generates 'A1'
tf.random.set_seed(1234)
print(tf.random.uniform([1])) # generates 'A1' again