Introduction to the course
This is the website for the
Inria Academy course
on the skrub package: it contains all the material
used for the course, including the datasets and exercises used during the session.
The material is available in two formats:
- a book, in
book/, with more detailed explanations;
- a set of slides, in
slides/, for instructor-led sessions.
Structure of the course
The course covers the main features of skrub, from data exploration to pipeline
construction. While skrub DataOps are a major feature of the package, they are
also expansive enough to deserve their own course, and as such only a short introduction
will be given here.
Each chapter includes a section that describes how a specific feature may assist
in building a machine learning pipeline, along with practical code examples, and
a quiz at the end.
The course is split in sections, which group relevant material together. Each
section is wrapped up by an exercise that covers what has been explained in the
section.
These exercises are stored in content/exercises as py files; running
pixi run materials converts them to Jupyter notebooks in content/notebooks
and puts the datasets in content/data, ready to be used.
Preparation and setup
Using Jupyterlite
The easiest way to work on the exercises is simply by using Jupyterlite: this
will create a notebook interface directly from the browser that allows to run the
exercises without needing to create a local environment.
Setting up a local environment
If you still want to work locally (for example, if you want to use your own IDE),
you can clone the GitHub repo
to have access to the exercises.
Finding the material
All the datasets are made available to the notebooks by cloning the repo; the
exercises are stored in content/exercises.
Run
pixi run materials
to generate the notebooks in content/notebooks (using
jupytext) and copy the datasets to
content/data.
Using pixi
The easiest way to set up the environment is by installing and
using pixi. Follow the platform-specific
instructions in the link to install pixi, then open a terminal window in the
root of the repository and run
pixi install
pixi run materials
pixi run lab
to create the environment, generate the notebooks and datasets, and start a
Jupyter lab instance.
Using pip
Create and activate the environment:
python -m venv skrub-tutorial
source skrub-tutorial/bin/activate
Install the required dependencies using the requirements.txt file:
pip install -r requirements.txt
Start the Jupyter lab instance from the root of the repository:
jupyter lab
Building the website locally
The website (landing page, book, slides and the JupyterLite instance) is built
with quarto and published on GitHub Pages by a GitHub
Actions workflow on every push to main.
To rebuild the full website locally, run
pixi run render
pixi run serve
then open http://localhost:8000 in a browser. The rendered site is placed in
the _build directory.