As cluster's basic operating system is quite old, therefore also its default python installation is ancient. That's why when working with python you should almost always start with loading suitable anaconda module and creating environment for your current python project. Otherwise you'll end up with a mess with broken python & libraries sooner or later. (You might still end up with it)
So start with either with:
- ..code-block::
- module load Miniconda3
# or
- ..code-block::
- module load local-anaconda
The first one is "mini-anaconda" specifically ment for creating your own environments and second one is locally maintained quite often updated and propably has some of useful modules already installed. But either way you'll start with more recent python and pip and with local-anaconda already have several python modules e.g. cuda and tensorflow already installed.
If/when you'll need to add some other python modules best way to proceed is to create your own anaconda environment "on top" of that:
- ..code-block::
- conda create -n my-env python numpy cuda ... # Let's activate newly created environment source activate my-env # now you can add either conda packages or plain python packages with conda install keras=2.4.3 pip install picos
You'll need to activate environment always before using it.
In case you want to deactivate it e.g. to create other one use source deactivate myenv
Your own environments are installed by default to your home-directory ~/.local so in case you get problems that's where you have start cleaning ;-)
Listing all environments known to conda: conda env list
And so on ...