Getting intel-tensorflow installation error in development environment

[Eriz11]

Hey all,

I have just cloned the master branch to use the commit that fixed the tensorflow.contrib errors regarding the cloud module.

However, when I install it in edit mode (pip install -e .) in a Linux desktop that doesn't have a GPU (i.e. I fall in the if sys.platform == "linux" or sys.platform == "linux2" > so the install requiresintel-tensorflow=1.13.1), although intel-tensorflow version 1.13.1 appears when doing conda list in the brand new conda environment that I create, I still get this error:

ERROR: rl-coach 1.0.0 requires intel-tensorflow==1.13.1, which is not installed

¿Could you please see if you also get this? @galnov @gal-leibovich

[galnov]

Other than the error message, do you see any problems running Coach after the installation process is complete?

[Eriz11]

Hey Gal,

Not really in the terms of the execution of the previous scripts that I have: I used Rainbow agent, DQN, DDPG... and all of them work properly without errors in the training.

I do not know about if in the backend something is going wrong, but it seems to be something related to the installation process, although that error seems to arise when the specific package is included in the requirements.txt or in the setup.py script, but in this case that is not the case so it shouldn't be screening it.

[AustinDeric]

this error breaks my install as well. Furthermore, i cannot find the release 1.13.1 for intel-tensorflow: https://github.com/Intel-tensorflow/tensorflow/releases. The closest is 1.13.2: https://github.com/Intel-tensorflow/tensorflow/releases/tag/v1.13.2-hardened

Would using 1.13.2 be acceptable?

[galnov]

1.13.1 can be found here: https://pypi.org/project/intel-tensorflow/1.13.1/