Anaconda first.
Create the environment:
conda create --name my-net python=3.6 scipy
That will create a virtual environment including scipy. So start the environment
source activate my-net
And when finished, use:
source deactivate
After this, your shell will probably be broken so it is a good idea to start a new one, but deactivating is recommended.
Of course, once the environment is started, you'll realize that you need more libraries. So to add more libraries,
conda install --name my-net keras
Finally, to remove an environment:
conda env remove --name simple-py
Next steps will be for virtual env/pip.
First, I installed virtual environment and pip from the centos repos, python34-pip and python34-virtualenv. (probably didn't need pip because virtual environment installs a local version it appears).
To create an environment and install packages.
virtualenv3 path/to/environment_name
source path/to/environment_name/activate
That is it, then to install packages.
pip install keras
And done. Also, there is a way to list all of the currently installed packages for migrating to a new system.
pip list