Python

We provide Python including a variety of packages (including numpy, scipy, pandas, scikit-learn, pytorch, jax, and other computational packages) through the Miniforge installer and the community-driven conda-forge channel.

Python versions¶

On our Linux servers, we provide Python 3.13. We can also help you access older versions of Python if needed, by using a Conda environment.

Note that in what follows we use mamba, a drop-in replacement for conda.

Installing packages with pip¶

To see what Python packages are available, invoke

mamba list

To install packages locally in your home directory (in ~/.local) use the --user flag to pip:

pip install --user package_to_install

It is possible to use conda install to install packages outside of a Conda environment, but we don’t recommend it as it can cause confusing interference between dependencies.

If you are using Conda environments you will generally want to install packages using conda/mamba.

Virtual environments (Conda environments and virtualenv)¶

Environments provide a way to manage Python packages (and with Conda environments even the version(s) of Python and other software) in a context that can be isolated and controlled. This allows one to more easily manage dependencies and provide for reproducibility.

Conda environments¶

Conda is a very popular tool for installing software, particularly widely used for creating Python environments and installing Python packages.

We provide extensive information on using Conda (or our preferred alternative, Mamba) on the SCF and elsewhere.

virtualenvs¶

virtualenv is an older approach to setting up Python environments.

To create a virtual env:

virtualenv --system-site-packages ~/path/for/your/env
source ~/path/for/your/env/bin/activate

At this point you can pip install packages (without --user):

pip install numpy

or do something more involved, such as:

git clone https://github.com/somerepo/somelibrary.git
cd somepackage
python setup.py install
# optionally, to delete source files
cd .. && rm -rf somepackage

When you want to escape out of this environment, run deactivate. To re-enter, run the source line as above.