Lightnews — Scholar-powered news

Remembering the order of the train_test_split

6

:probabl. @probabl.ai · Feb 6

The page for the `train_test_split` function is the most visited page on the scikit-learn documention. It's because people always forget the order of the output.

So we figured it might make sense to make a video that helps you remember:
youtu.be/wb4bKSHiDik

The ups and downs of stratified cross-validation

1

:probabl. @probabl.ai · Jan 30

When you cross-validate, you're usually using stratified sampling. It's a default that comes with scikit-learn, and it certainly has an upside ... but also a downside.

We try to explain why in our latest whiteboard video on YouTube.
www.youtube.com/watch?v=EpK9...

Introducing the EstimatorReport

3

:probabl. @probabl.ai · Jan 28

We introduced a new tool last week that helps you immediately grab all the right metrics and charts for your pipeline. It comes with many sensible defaults and can be customised too.

If you're keen to save some time, feel free to check the demo on YT:
youtu.be/2rjkrR7CjHc

Exploring prefect for scheduling data workloads

2 5

:probabl. @probabl.ai · Jan 24

We will be going live in a bit, if you're keen to learn about prefect, feel free to join!

youtube.com/live/hkm5Ofi...

Playing with the classification report

youtube.com

1

:probabl. @probabl.ai · Jan 23

We have a new whiteboard video! This one is all about the classification report and we explain how it works in detail by leveraging a custom widget to change some of the values by hand.

You can watch it on our YT channel here:
youtu.be/765qaIk30Rs

Introducing the EstimatorReport

2

:probabl. @probabl.ai · Jan 22

Demo time! 2mn of your time to get a flavor of skore 😉 youtu.be/2rjkrR7CjHc

GitHub - probabl-ai/skore: the scikit-learn sidekick

1 10

:probabl. @probabl.ai · Jan 22

🚀 Meet skore—the @scikit-learn.org sidekick!

🔹 Offers guidance on modeling
🔹 Automated reports & key metrics
💡 Built by scikit-learn maintainers. Open-source & ready to use!

Try it 👉 tinyurl.com/bdhszwtn

the scikit-learn sidekick. Contribute to probabl-ai/skore development by creating an account on GitHub.

tinyurl.com

1 7 14

:probabl. @probabl.ai · Jan 21

Another week, another livestream.

This time we will have a look at prefect. It's a modern scheduling tool for Python that seems to have some nice features for ML training. Join us here on Friday:

youtube.com/live/hkm5Ofi...

Exploring prefect for scheduling data workloads

More Python parallel stuff with joblib-modal

youtube.com

5

:probabl. @probabl.ai · Jan 17

We are going live in a bit.

If you're keen to see some parallel techniques in Python, or want to learn about Modal/joblib, this one is for you.

www.youtube.com/live/0SXyIcs...

Share your videos with friends, family, and the world

2

Reposted by :probabl.

Yann Lechelle @ylechelle.bsky.social · Jan 17

Looking forward to speaking on the #OpenSourceAI panel at the sold out EU Open Source Policy Summit by @openforumeurope.org in two weeks. Rather than discussing the definition, we’ll focus on how #openness contributes to European competitiveness—or perhaps, how it might help Europe catch up.

3 10

:probabl. @probabl.ai · Jan 14

The livestream this week will be about joblib and some big time cloud parallelism that it can support these days via a Modal integration.

If you want to learn about Modal or joblib, this should be a super fun one.

www.youtube.com/live/0SXyIcs...

YouTube

Share your videos with friends, family, and the world

:probabl. @probabl.ai · Jan 10

We're going live in a bit with a new livestream. In this one we're giving ModernBERT a spin. It is very early days for the project, but there are some interesting milestones already.

Join here at 12:30 CET:
youtube.com/live/ZWo6Q85...

YouTube

youtube.com

2 3

:probabl. @probabl.ai · Jan 9

The first whiteboard video of the year talks about dummy models. The name might not imply it, but using these "dumb" models can be a very smart thing to do at times.

youtu.be/0553mUt3O-E

Dumb models can be very smart

When precision equals recall

3

:probabl. @probabl.ai · Jan 7

We are back with livestreams this week and this Friday we will explore ModernBERT, a new pragmatic BERT model. It is very early for the project, which seems like a fun time to dip our toes!

Join us on Friday during lunch here:
youtube.com/live/ZWo6Q85...

11

Reposted by :probabl.

Yann Lechelle @ylechelle.bsky.social · Dec 25

Rather than predictions, here are my professional wishes for 2025—shaped by a year of profound learning, building mission-driven @probabl.ai and data science commons with and around @scikit-learn.org, and engaging with brilliant minds, both enthusiastic and concerned about AI: #openness

4 15

:probabl. @probabl.ai · Dec 23

Our Probablrs Starter Pack is up! go.bsky.app/JZXryZd

2

:probabl. @probabl.ai · Dec 19

Our final whiteboard video of the year is about a detail, a tiny tiny detail, that relates to the precision and recall metrics. These two metrics are typically different, but sometimes they are equal.

And something interesting is happening when that is the case!
youtu.be/Zn2xrf4ijSY

Ask me Anything[tm] and a certification review!

1 6

:probabl. @probabl.ai · Dec 17

We are going offline during the holiday weeks, which is also why we are going to have a special edition of the livestream to wrap up the year. Feel free to ask us anything about ML as we also review some certification preparations.

www.youtube.com/watch?v=f9G7...

Making a recommender by just using Polars!

3

:probabl. @probabl.ai · Dec 16

Our new handle is @probabl.ai :)

2 12

:probabl. @probabl.ai · Dec 16

done

:probabl. @probabl.ai · Dec 13

Going live in a bit!

www.youtube.com/watch?v=U3Fi...

Precision, recall and F1-score

4

:probabl. @probabl.ai · Dec 12

We're going to dive into some metrics in the next few whiteboard videos. After all, if you are going to use them to judge machine learning models, it is really important to have a good intuition for them.

Let's start with precision, recall and F1:
youtu.be/3M2Gmuh5mtI