How do I know which version of the Keras API is implemented in tf.keras?

Question

In the standard implementation of Keras, one can get the API version using keras.__version__.

However, there is no tf.keras.__version__.

So how do I check the version of the Keras API implemented in tf.keras?

score 5 · Accepted Answer · edited Oct 15 '19 at 17:42

5

You can simply run this code

from tensorflow.python import keras
print(keras.__version__)

which is compatible with TF v1.8.0.

edited Oct 15 '19 at 17:42

nbro

answered Jul 05 '18 at 19:58

Abhishek Sehgal

On TF 1.8 it returns `AttributeError: module 'tensorflow.tools.api.generator.api.keras' has no attribute '__version__'`. Which version of TF are you running? – P-Gn Jul 06 '18 at 06:03
You're right. In 1.8.0, I checked and it didn't work, but this was working in previous versions. So I tried `from tensorflow.python import keras; print(keras.__version__);` It printed `2.1.5-tf` – Abhishek Sehgal Jul 06 '18 at 15:26
Yes, this one works — and is shorter that my version. If you edit your solution I will accept it. – P-Gn Jul 07 '18 at 08:32

score 2 · Answer 2 · answered Feb 28 '19 at 19:07

2

This has recently been updated:

>>> import tensorflow as tf
>>> print(tf.keras.__version__)
2.1.6-tf
>>> print(tf.__version__)
1.12.0

answered Feb 28 '19 at 19:07

user3076252

score 1 · Answer 3 · answered Jul 05 '18 at 15:19

You can retrieve the version of Keras implemented in tf.keras using

from tensorflow.python.keras._impl.keras import __version__ as tf_keras_version
print(tf_keras_version)
# 2.1.5-tf (in TF 1.8)

It seems that from TF 1.9 on, it will actually be accessible as tf.keras.__version__, as it should.

3 Answers3