What is compatible Keras and Tensorflow versions?

[mrgloom]

What is compatible Keras and Tensorflow versions assuming hyperas is installed via pip install hyperas using anaconda3? Maybe it's better to include requirements.txt in repo?

[marcoleewow]

I am having problems with Keras 2.0.4 and tensorflow 1.1.0. After I pip install hyperas, when I import the library it gives NameError: global name 'exc' is not defined error. Hope there will be an update soon.

[maxpumperla]

still one week on vacation. will get back to this. thanks for understanding

[mrgloom]

Seems it work for me on:

./pip show hyperas
Name: hyperas
Version: 0.4
./pip show keras
Name: Keras
Version: 2.0.5
./pip show tensorflow-gpu
Name: tensorflow-gpu
Version: 1.2.1

Can you consider upgrade?

[marcoleewow]

Sorry I figured out my problem now, I was using elephas docker and I noticed that I did not install hyperas properly. I will add a new issue in elephas instead, disregard my comment before.

[impredicative]

@mrgloom requirements.txt is generally meant for standalone applications only, not for imported packages e.g. hyperas. Note that setup.py already documents the version-agnostic requirements as is normal. Even so, if requirements.txt had to forcibly be used, setup.py would have to read from it - this doesn't benefit much.

As for keras and tensorflow version restrictions, defining them would restrict the usability of hyperas a bit too much. It's supposed to work with the latest stable versions. It may still make sense to define the minimum version requirements, but not the maximum.

[dineepthomas]

@mrgloom tensorflow==1.4.1 and keras==2.0.8 works for both cpu and gpu.