setup.py and tensorflow with/without gpu about tensorforce

[michaelschaarschmidt]

Good point, we will get to this, thanks!

[pedronahum]

Hi @krfricke

The current setup.py will fail to recognize if tensorflow-gpu is already installed. In fact, it would download and install the cpu version.

If I make the changes as shown below, would allow the following installation options:

--Assumes tensorflow or tensorflow-gpu installed
pip install tensorforce -e .

--Installs with tensorflow-gpu requirement
pip install tensorforce[tf_gpu] -e .

--Installs with tensorflow (cpu) requirement
pip install tensorforce[tf] -e .

setup(name='tensorforce',
version='0.2',
description='Reinforcement learning for TensorFlow',
url='http://github.com/reinforceio/tensorforce',
author='reinforce.io',
author_email='[email protected]',
license='Apache 2.0',
packages=['tensorforce'],
install_requires=[
'numpy',
'six',
'gym>=0.7.4',
'universe>=0.21.3',
'scipy',
'pillow',
'pytest'
],
setup_requires=[
'numpy',
'recommonmark'
],
extras_require={
'tf': ['tensorflow>=1.2.0'],
'tf_gpu': ['tensorflow-gpu>=1.2.0'],
},
zip_safe=False)

I guess I am very motivated to use the library :o)

Thanks,

Pedro

[krfricke]

I will add some instructions to the readme. Thanks for your help, Pedro!

[krfricke]

Sorry, I forgot to remove tensorflow from the install_requires list when I added the extras_require dict. I pushed an update in f4fa33c. I currently can't confirm it fixes the issue as I don't have tensorflow-gpu installed. It would be great if you could check if the commit solves the issue.

[pedronahum]

Hi,

Confirmed! And the following 3 options are still valid:

--Assumes tensorflow or tensorflow-gpu installed
pip install tensorforce -e .

--Installs with tensorflow-gpu requirement
pip install tensorforce[tf_gpu] -e .

--Installs with tensorflow (cpu) requirement
pip install tensorforce[tf] -e .

Maybe it is worth to update the installation instructions as well...

Thanks!