Difficult to use tensorflow-addons with tf-nightly builds

[ghop02]

It's a bit tricky to use the tf-nightly builds and have a dependency on tensorflow addons because of the pinned tensorflow version.

Is there a way to have the setup.py file for the nightly builds be slightly more lenient as to the version? It's a bit tricky because tf-nightly is an entirely different package.

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
• TensorFlow version and how it was installed (source or binary): tensorflow nightly (pip)
• TensorFlow-Addons version and how it was installed (source or binary): (0.7.1)
• Python version: 3.7

Code to reproduce the issue

pip install tf-nightly
pip install tfa-nightly (or tensorflow-addons)
# installs tensorflow 2.1.0 which downgrades gast

[seanpmorgan]

Hi @ghop02! Thanks for reporting. So we should have a soon incoming solution to this which is to remove the hard installation dependencies:
#864
#952

I plan to work on this later today when I have some time to rework the builds. If you need to unblock today you can pip install tfa-nightly --no-deps

[ghop02]

Great, figured there was something in the works!

And never knew about --no-deps, will probably abuse this now for many carefully constructed brittle python environments 😄

[seanpmorgan]

TFA Nightlies starting tomorrow will have no TF pip dependency. 0.8 release is upcoming and will also be dependency free