Organizer

• Adeline Fermanian, Head of Research at Califrais

Workshop Introduction:

Califrais is a company that develops machine learning algorithms and new technologies to optimize large-scale food flows. They specialize in finding innovative solutions to improve the efficiency and effectiveness of supply chain management in order to decarbonize the global food supply chain, reducing food waste and CO2 emissions.

Notably, Califrais operates on Paris Rungis market, the largest fresh produce market in the world. They also work with academic institutions (Sorbonne Université, CNRS, Université Paris Cité…) through a public-private research collaboration structure, the LabCom LOPF (Large-scale Optimization of Product Flows). Their research topics are at the intersection of machine learning, logistics optimization and ecology.

In this workshop, we will have the opportunity to work on Califrais’ data, representing 2 years of daily sales on hundreds of fresh products. Our objective will be to build univariate time-series forecasting models at a 1-month horizon to predict demand on all fresh products.

Workshop Summary:

In this workshop, we will:

1. Introduce time-series modelling as well several best practices when working on them.
2. Visualize and train forecasting models on the time-series.
3. Compare results and methods with other participants.

If you ever wondered how to get started with time-series modelling, this workshop is for you!

Come and benefit from the experience of Califrais on this domain.

Organizers

• Raphael Vienne, Head of AI at datacraft
• Rémy Gasmi, Data Scientist Intern at datacraft

Workshop introduction:

Scraping has been more and more recognized since LLMs became a thing, as these models rely on several petabytes of internet data for pre-training, that were extracted from web crawlers.

Every year, the internet produces tons of extremely valuable data. Some individuals might be interested in either collecting relevant data from the internet automatically, or even automate some actions online.

Both of these considerations can be done with scraping.

In this workshop, we will try to introduce participants to scraping, as well as discussing legal considerations regarding this practice.

Workshop summary:

In this workshop, we will:

1. Introduce scraping libraries as well as legal considerations regarding scraping (when not to scrape).
2. Start scraping on a simple example (extracting information from a wiki).
3. Carry out a more complex scraping pipeline (scrape datacraft agenda and incoming events).
4. Finally, let participants build their own scraping project (on the website of their choice).

If you thought of automating an online task once in a while, or if you’re just curious about scraping with python, this workshop is for you!

Come and benefit from the experience of our team on this domain.

• Raphael Vienne, Head of AI at datacraft
• Rémy Gasmi, Data Scientist Intern at datacraft

Workshop introduction:

Data processing is a key part of a data scientist’s day to day job. Today, we consider that most data scientists spend more time processing, and visualizing data than building models out of it. Another key finding is that better downstream performance is often yielded from data quality and robustness of data pipelines, rather than from architectural improvements.

For several years, pandas has shown to be the go-to open-source python library for single-node data processing. However, its creator, Wes McKinney, published in 2017 a blog post entitled: “Apache Arrow and the 10 things I hate about pandas” where he goes through several design choices that were made during the development of pandas, and how he would do them differently, had he had the opportunity to do things differently.

From this idea, polars was born.

Polars started out as a hobby project in 2020, but quickly gained traction within the open source community. Many developers were searching for an easy-to-use DataFrame library that was performant at the same time, and Polars set out to fill this void. The community grew fast as many contributors came in from various backgrounds and programming languages.

Today, polars is rapidly evolving and community adherence is very strong. The library is evolving at a pace where it could outgrow pandas (in terms of github stars) within several years.

Workshop summary:

The goal of this workshop is to introduce data scientists to the polars library and provide first examples to become familiar with it.

In this workshop, we will:

1. Briefly introduce polars and the design choices associated with the library.
2. Work our way through the documentation and basic functions / objects as a starter.
3. Translate complex pandas pipelines to polars.
4. Evaluate the gain in performance associated to various tasks that a data scientist can work on.

Whether you heard of polars or not, let us convince you that this library is not something you want to miss.

Come and benefit from the experience of our team on this library.