CODEWITHSUNDEEP
machine-learningtensorflowcomputer-vision
j35t3r
j35t3r

Reputation: 1523

Is there a difference between activations functions in tensorflow? tf.nn.tanh vs tf.tanh

I want to setup a neuronal network and I am asking myself if there is a difference between those two functions?

tf.nn.tanh vs tf.tanh

Upvotes: 2

Views: 6096

Answers (4)

bogatron
bogatron

Reputation: 19169

Easy enough to confirm they are the same:

In [1]: import tensorflow as tf

In [2]: tf.nn.tanh
Out[2]: <function tensorflow.python.ops.math_ops.tanh>

In [3]: tf.tanh
Out[3]: <function tensorflow.python.ops.math_ops.tanh>

In [4]: tf.nn.tanh == tf.tanh
Out[4]: True

In [5]: tf.__version__
Out[5]: '0.11.0rc1'

Upvotes: 5

Rodolfo Bonnin
Rodolfo Bonnin

Reputation: 81

No there isn't any difference.

The availability of both is probably due to the library evolving and still changing its API, being still in an initial state of maturing. We could expect the library to avoid those fundamental duplications in the future when the main API is finally set (I'd expect so towards 2.0).

Upvotes: 2

nessuno
nessuno

Reputation: 27042

No, there's no difference.

In the tensorflow/tensorflow/python/ops/nn.py file (that's where tf.nn is defined) we can find the definition of tanh:

from tensorflow.python.ops.math_ops import tanh

also, there's this TODO here

# TODO(cwhipkey): sigmoid and tanh should not be exposed from tf.nn.

Thus, probably tanh will be removed from the tf.nn package.

Hence tf.tanh (that's defined here) is the one to use.

Upvotes: 4

P-Gn
P-Gn

Reputation: 24581

They are the exact same alias to tensorflow.python.ops.math_ops.tanh.

Same thing goes for tf.sigmoid and tf.nn.sigmoid.

Upvotes: 2

Related Questions