Story of scikit-learn's life

We merged my most wanted scikit-learn feature: a pair of estimators to modify the decision threshold of a binary classifier. Nice use case: tinyurl.com/5exejdz5; and avoid some traps: tinyurl.com/3cxm89s8. Thanks to all reviewers and contributors.

That's a really big deal, IMHO: this feature is not 'impressive', but it's the kind of things that makes the success of a data-science project.

Congratulations to Guillaume and everyone involved!

The hidden mechanism behind AI regulations:

1. Claim that AI can do everything .
2. Raise tons of money from investors.
3. Tell governments that AI is very dangerous and that open source AI should be regulated out of existence.
....
4. Profits!

May 1st toy project: update the project template repository for scikit-learn-contrib projects: github.com/scikit-learn-c…. Amazing to see that in a couple of hours, you now get started with CIs and doc deployment thanks to GitHub Actions and pixi from prefix.dev

I am really excited to be part of this project led by Sayash Kapoor and Arvind Narayanan to help improve practices in machine-learning based science. science.org/doi/10.1126/sc…

📣Early bird tickets for PyData Paris 2024 are out!
🗓Sep 25-26 at citedessciences
🌟Keynotes:
- Mistral AI's Arthur Mensch
- Thoughtworks' Katharine Jarmul
- :probabl.'s Olivier Grisel
- Hugging Face's Merve Noyan
🤝Connect with the best in open-source scientific computing!

I'll be diving into sparse data this friday on the :probabl. livestream. If you're curious to see some tricks that keep memory requirements low ...

youtube.com/live/u_cFkXFGj…

Google, who massively benefits from Python, stops investing in its maintenance and improvement!? 😱

Is that true, Jeff Dean (@🏡)? How is that justified?
Digital commons, as OSS, work only if all pitch in, in particular the privileged.

The StandardScaler in scikit-learn sounds like an easy thing to build until you realize that it also needs to support sparse data, sample weights *and* online learning. Not to mention numeric stability.

The StandardScaler, it turns out, is not standard!

youtube.com/watch?v=VZxqqH…

🍿 I had the great pleasure of chatting with Gael Varoquaux @GaelVaroquaux.bsky.social, a co-founder/OG core dev of scikit-learn, recently in Paris for Outerbounds TV 📺 . Gaël

* Started scikit-learn, is a scientist at Inria, and is helping to spin out :probabl.! 🎲
* Loves using science and…

Today I presented at PyConDE & PyData Berlin 'A Retrieval Augmented Generation system to query the scikit-learn documentation'. The GitHub repository is available at github.com/glemaitre/skle… and the slides at tinyurl.com/2s3c5shn

This is a very important point: reviewers too often ask for experiments that are beyond what many labs can afford.

On the one hand, solid experiments make better scientific evidence.
On the other hand, bruteforcing is not sustainable and leaves behind non-privileged actors

'Boring AI' technology like scikit-learn is your antidote to 'hype AI'! Hugo Bowne-Anderson from Outerbounds says it best. To dig deep into your data and improve your company performance, there is nothing like a little a fit-predict routine to save the day. Actually exciting😉

Do you love open source, open science, research and education? Interested in connecting AI with science and society, in an incredible intellectual community? This weekend is a great time to update your CV and apply to the BIDS - Berkeley Institute for Data Science Executive Director position!

I think we should be building systems that complement people; systems that do well the things that people do poorly; systems that make individuals and organizations more effective and more humane. 3/

Listening to Mark Cuban speak about 'scale capture' in health: large actors, hard to circumvent because of a complex system that they create), I can't help to wonder if there is a parallel with modern cloud-based software & AI:
A huge invisible added cost?

podcasts.google.com/feed/aHR0cHM6L…

Dataiku delivers 'Everyday AI' since 2013, helping the modern enterprise leverage data, machine learning and AI at scale. Florian Douetteau is the quiet force that led this wonderful company scale from 0 to 1 in just a decade, becoming a beacon of excellence in enterprise software.