Can others reproduce
your results?
Using uv to manage dependencies
in a Python data analysis project

James Thomas
j.thomas@bristol.ac.uk

2025-06-05

Outline

• What is uv? (and pixi?)
• Why might you want to use them?
• How do you configure them?
• Some examples

Note:

• I am not the developer of these tools
• You may well know more than me – tips welcome 😁

What does uv do?

Similar to venv, pip, pip-tools, poetry, …

• install python and PyPI packages
• manage a virtual environment (.venv) and lockfile

• First released in February 2024, initially with functionality of venv, pip and pip-tools.
• But allows you to install different versions of the Python interpreter, if you don’t have them on your system already.
• Makes it good for bootstrapping where someone doesn’t currently have Python.
• Support for projects was introduced later in the year, keeping the installed packages up-to-date with a lockfile (more on this later).
• Development is fast-paced. Current version (0.6) first released February 2025 and most recent release (0.6.11) was 2 days ago.

and:

• cheap disposable environments (jupyter lab, pytest)
• inline script metadata (recreating bugs, demos)
• globally-accessible tools with uvx (ruff, visidata)
• cross-platform, easy to install, fast

• With uv, virtual environments are cheap to create. So you can temporarily add packages to the environment.
• For one-off scripts, you can specify the dependencies at the top of the script file (PEP 723).
• You don’t even have to install global tools - think pipx - they are downloaded and cached on demand.
• (More on these later)

What does pixi do?

Similar to uv, …

• install conda packages (python or R) or PyPI packages
• manage a virtual environment (.pixi) and lockfile

and:

• task runner
• globally-accessible tools (sync between devices)
• cross-platform, easy to install, fast

• Pixi can be though of as the uv equivalent for the conda ecosystem.
• From the developers of mamba/micromamba.
• Advise check it out - remainder of this talk will focus on uv, but concepts are similar.

Why properly managing dependencies
is important (a motivating example using pip)

• Promised an example (pip_demo). Will show what can go wrong if you use pip on its own.
• Start with a skeleton system with Python 3.10 installed and ls -A a venv created
• cat my_script.py that performs simple hypothethis test using scipy
• cat requirements.txt but can see this project was created a while ago, versions are old
• source .venv/bin/activate
• pip install -r requirements.txt (numpy 1.21.6, scipy 1.7.3)
• python my_script.py
• Now want to do some plotting pip install matplotlib (3.10.0) which upgrades numpy to 2.2.2 (note the warning)
• Breaks python my_script.py
• nano requirements.txt and add matplotlib
• pip install -r requirements.txt works but needs to download multiple versions to find the right one (3.8.4)

What about the workflow with uv?

• (uv_demo)
• ls -A note no venv - python not even installed
• https://docs.astral.sh/uv/ and curl -LsSf https://astral.sh/uv/install.sh | sh
• uv init --python 3.10 and note ls -A the contents (can remove hello.py)
• uv run python -V to verify version (note downloads python and creates venv)
• uv add numpy==1.21.6 scipy==1.7.3 (as before)
• uv run my_script.py (note python is assumed and no need to activate)
• uv add matplotlib
• uv run my_script.py (works and uses 3.8.4, same as before)
• What has happened?
• cat pyproject.toml and cat .python-version
• cat uv.lock another toml file with all the matching versions of the packages installed, plus hashes.
• (version control all of these - automatically up-to-date)

Configuring uv using pyproject.toml

• Can make changes to pyproject.toml and .python-version using a text editor or using uv. Shouldn’t edit uv.lock by hand.
• Showing this both ways
• uv python list
• uv python pin 3.13 but this won’t work as the packages are too old
• nano pyproject.toml and change version constraints to >=
• uv sync --upgrade to update all packages to latest versions possible, given the constraints
• uv run my_script.py

Applying this in a data science context

• (uv_lib_demo)
• Switch to example for uv init --package but have made some other changes
• tree to show layout, come back to notebooks, cat src/my_project/probability.py
• cat pyproject.toml including build system and project scripts
• uv run test-coin
• To look at notebooks uv run --with jupyter jupyter lab including importing and autoreloading

Other tips

• Scripts with inline metadata (PEP 723)
  • cd ../uv_script_demo
  • cat my_script.py
  • uv run my_script.py
  • run again
• Global tools - if you recall, like pipx
  • uvx ruff format my_script.py
  • cat my_script.py to show changes
  • useful for running tools on PyPI without having to formally install them

Wrapping up

• Find out more: 🔗 docs.astral.sh/uv (and 🔗 pixi.sh)

• Use for new projects

  • uv if PyPI (pixi if Conda)
  • package in src/
  • commands uv run test-coin

• When working with someone
  new to Python

  • uv run ... could be in a script

• Don’t use pip/conda by themselves

James Thomas
j.thomas@bristol.ac.uk

jatonline.github.io/managing-
dependencies-using-uv-and-pixi

Bonus material - pixi

• (pixi_demo)
• cat pyproject.toml
• cat pixi.lock
• pixi run test-coin
• pixi run format
• pixi run lab