Does TensorFlow 1.9 support Python 3.7

Question

I'm hesitating whether to downgrade to Python 3.6 or install a new version of TensorFlow.

Does TensorFlow 1.9 support Python 3.7?

Answer 1

I was able to install Tensorflow 1.12.0 with Python 3.7 on MacOS, with the following command.

sudo python3 -m pip install --upgrade https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.12.0-py3-none-any.whl 

Answer 2

Not yet. It seems there are some variables named "async", which has become a keyword in 3.7. Shouldn't be too difficult to fix, but still a problem.

Source: https://github.com/tensorflow/tensorflow/issues/20444

Answer 3

Probably not yet.

First of all, you will probably get a SyntaxError: invalid syntax because some parameters and variables at the pywrap_tensorflow_internal.py module have the name async which is a reserved keyword in Python 3.7.0 (see also this post).

However, you can solve this by simply changing the name of all these (Ctrl + R and Replace All) from async to for example async1.

The problem is that then you will probably get a ImportError: No module named '_pywrap_tensorflow_internal' which perhaps may be fixed for any previous version before Python 3.7.0 as this post on StackOverflow suggests but personally I could not fix it on my laptop with Python 3.7.0 .

Consequently, I simply downgraded to Python 3.6.6 and now Tensorflow is working fine.