Stable-baselines: Tensorflow requirement not satisfied but installation succeeds
I tried installing stable-baselines in a virtualenv in python 3.7.7 and Ubuntu 18.04:
pip install stable-baselines
Although there is no version of TensorFlow installed in that environment, the setup doesn't collect and install it and proceeds with:
Successfully installed stable-baselines-2.10.0
How is that?
mhtb32
>All comments
How is that?
That's an issue of pip and tensorflow in fact...
If you want the correct tensorflow version to be installed at the same time, use pip install git+https://github.com/hill-a/stable-baselines (it will install the version from master and handle tensorflow dependency properly).
Because tensorflow could not handle properly cpu vs gpu installation, we had to put a condition in the setup.py. However, after packaging stable-baselines into a pip package, this condition is no longer checked... As a result, you need either to install from source (pip install git+https://github.com/hill-a/stable-baselines) or install the correct tensorflow version manually.
araffin
on 11 May 2020
Related issues
SerialIterator
路
3Comments
maystroh
路
3Comments
JankyOo
路
3Comments
ktattan
路
3Comments
matthew-hsr
路
3Comments
Most helpful comment
That's an issue of pip and tensorflow in fact...
If you want the correct tensorflow version to be installed at the same time, use
pip install git+https://github.com/hill-a/stable-baselines(it will install the version from master and handle tensorflow dependency properly).Because tensorflow could not handle properly cpu vs gpu installation, we had to put a condition in the
setup.py. However, after packaging stable-baselines into a pip package, this condition is no longer checked... As a result, you need either to install from source (pip install git+https://github.com/hill-a/stable-baselines) or install the correct tensorflow version manually.