Getting Started
Binaries
We provide the following modules with Python binaries:
lang/Python
lang/Anaconda3
Please contact pc2-support@uni-paderborn.de if you need other binaries or distributions.
Installing packages
Environments
We highly recommend to install python packages in local environment rather than in a global package depot. Common choices are conda environments or venv environments (see here for a non-conda overview).
Where to install packages?
The recommended place for Python packages/environments is PC2PFS, i.e. under /scratch
. Note that /scratch
is not shared between Noctua 1 and Noctua 2 and you will thus need to maintain separate package depots/environments for each system. This is the recommended approach because both systems have very different architectures (e.g. Intel vs AMD CPUs) such that binaries may easily be incompatible.
Note that HOME is not a good choice for storing Python packages since 1) it is not designed for parallel use, 2) will very likely lead to quota issues (HOME is limited to 5GB), and 3) causes unnecessary backups of reproducible data.
Changing global install location
Pip
Without intervention, and if your’e not using local environments, pip install -U <package>
will install packages into $HOME/.local
, which is discouraged for the reasons mentioned above. To change the default install location, we recommend to put something like the following into your .bashrc
:
export PYTHONUSERBASE=/scratch/<group>/<username>/.local
Conda
By default, conda will install package data and environments into $HOME/.conda
. This should be avoided for the reasons mentioned above. To change the global default, execute the following commands, which will set the corresponding paths in your $HOME/.condarc
:
conda config --add envs_dirs /scratch/<group>/<username>/.conda/envs conda config --add pkgs_dirs /scratch/<group>/<username>/.conda/pkgs