Repository for the tutorial session "Performant, scientific computation in Python and Rust" at PyCon/PyData Berlin 2024.

To the participants of the tutorial session:

All stages of the tutorial can be reproduced by checking out different branches of this repository and the slides are written in HTML and will be hosted on GitHub pages shortly before the session so that you can see them in our browser.

Note: If you want to reproduce the tutorial on your device, please follow the installation steps described below prior to the session to avoid longer delays during at the beginning. Thank you very much!

I use uv for dependency handling but you can use pip-tools or plain pip / virtual environments instead.

Uses a modern tech stack based with a lot of (opinionated) choices:

• Python (^3.12)

  • Pyenv
  • uv
  • PoeThePoet (task runner)
  • Ruff (extremely fast linter)
  • Mypy (type checker)
  • Jupyterlab, matplotlib (--extra lab)
  • Sphinx, pydata theme, and markdown via myst-parser) (--extra docs)
  • numpy, scipy, sklearn
  • pre-commit hooks

• Rust

  • ndarray (equivalent of numpy)
  • PyO3 (Python bindings)
  • cargo-show-asm

• Development environment (recommendations)

  • Visual Studio Code or VS Codium
  • Fira Code Font (e.g., sudo apt install fonts-firacode)

The repository is structured as follows:

• python sources for the pure python project

• rust sources for the optimized code in rust

• bindings sources for wrapping the rust code for python.

  This is separated from the rust crate as this is required for inline testing

• pyproject.toml, requirements.txt The definition of the python project and the locked virtual environment

• Cargo.toml, Cargo.lock: The definition of the workspace and locked dependencies

I recommend installing with PyEnv.

curl https://pyenv.run | bash
pyenv install -l # choose a recent version
pyenv install 3.12.0

Alternatively, you can use the system's version of course. On macOS using brew is recommended, while on Debian-based Linux distributions, run

sudo apt install python3.12

I have little experience with Windows systems, but you should probably use WSL2 there.

First, install uv (or follow instructions on the project page)

curl -LsSf https://astral.sh/uv/install.sh | sh

Create environment and install the package

pyenv shell 3.12 # optional to select version when using pyenv
uv venv
# The next line needs only be called if dependencies are altered
uv pip compile pyproject.toml --all-extras -o requirements.txt
uv pip sync requirements.txt
. ./venv/bin/activate # Optional to activate the environment

Alternatively, you should be able to achieve the same with plain pip

python3.12 -m venv .venv
. ./venv/bin/activate # Optional to activate the environment
pip install -r requirements.txt

You should select the newly created virtual environment in your development environment.

In pyproject.toml there are multiple tasks defined that are available once you have the virtual environment build and activated:

poe # will list all commands
poe lint # runs ruff to fix all auto-fixable issues
poe typing # runs the mypy type checker
poe fmt # runs ruff to format the source files
poe test # runs test and coverage
poe all # runs all of the above
poe install-kernel # makes the environment available in jupyter
poe uninstall-kernel # removes the associated jupyter kernel
poe verify # Checks whether cuda works as expected
poe docs # Builds the docs
poe lab # starts a jupyter service

poe docs
firefox docs/build/html/index.html # or similar

Activate the virtual environment and run

pre-commit install

Then, all modified sources will be checked for formatting and all tests will be on every commit unless if called with the -n (no-verify) switch. The virtual environment needs to be activated for this. If you run into problems with your IDE, please install pre-commit globally for your user.

After calling the following command, changes in Python are effective immediately from now on (restarting the interpreter might be required). Changes in Rust require a rebuild with the same command.

uv pip install -e . -v -U # for an editable install
# or
poe maturin               # for building a wheel

First install Rust. This is usually done with the rustup tool that manages Rust installations and distribution of tools. On a Unix-like OS, run

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
rustup update

You can check your installation by building the project. The command used for this is cargo a build system (comparable to poetry or rye in the Python world):

cargo build

You should also install the following add-ons to the cargo build system:

cargo install cargo-criterion
cargo install cargo-show-asm