Fifteen years ago, I got my dream job writing for MLSsoccer. Last week, that came to an end.
I'm off on my own, and it's time to build something new โ with a little help from all of you.
tacticsfreezone.ghost.io/an-end-to-my...
En route to Houston for @americansoccerinsights.com summit. Really looking forward to catching up with some old faces.
Bar chart Title: Best sources for women's soccer data Subtitle: Women's competitions with xG Data | 8 comps surveyed Chart shows # of comps on y-axis. Only FBref scored 8/8. Next closest is Fotmob at 4.
In the wake of FBref's defacto death, I saw a lot of posts suggesting alternate data sources. Unfortunately, many of them focus exclusively on men's soccer. To measure what options remain for woso, I conducted a survey of the most popular FBref alternatives. What I found is disappointing:
๐งต
The best thing about LLMs in data science is that nobody forgets to label their axes anymore.
We have an opening on our Data Analytics team at US Soccer for a Performance Data Analyst. This person should have strong foundational skills in data analysis (python preferred) as well as experience in the Sport Science/human performance space (any sport)
recruiting.paylocity.com/Recruiting/J...
When your step count is larger than your sample size.
Ahead of the election, I walked 65 km around my electoral district and took a picture of every political lawn sign.
www.centralwinger.com/p/getting-st...
I have the same suspicion, but you're better qualified to answer that question!
So if Distance is flat, and Sprints are up, that means ... Walking must also be up?
There has to be some sort of reciprocal change to the ensure balance.
I'm having a very difficult time believing this data, especially after the inflation of stoppage time added minutes.
Weโre hiring for a very unique Data Coordinator contracting role.
jobs.smartrecruiters.com/MLSE3/744000...
This is great! Very intuitive. I suppose pooling will have this effect with any sort of aggregation along an axis? Pretty useful.
We're talking football! Well kinda. @centralwinger.com asked if anyone was interested in explaining The Zoo model, so I figured I'd give it a go.
thebucketless.wordpress.com/2025/04/01/e...
Not silly! I donโt think thereโs anything wrong with eyeballing it (which is what I did here).
Itโs difficult to โaverageโ a set of points because of the same order-invariance issue.
Iโve found success with something called a Wasserstein Barycenter.
First blog post in a while.
Tracking data research probably wasnโt worth it until recently. And some unexpected utility of the Hungarian algorithm.
www.centralwinger.com/p/tracking-d...
Lucasfilm should have hired @mikeduncan.bsky.social to do a Revolutions-style accompanying podcast for Andor. Such a missed opportunity.
Is there a conceptual issue with performing a bayesian update on xG measurements after a shot? The outcome is pretty useful information, and it's a mathematically valid approach.
I assume someone has already tried this?
This is awesome!
That's what I've been playing with! But would prefer some of your feature engineering out of the box!
What Iโd really love is if unravelsports could be used to make node-level predictions instead of just frame-level.
One of the first things I want to try with this is plugging in your GNN model.
Big thanks to @probberechts.bsky.social โ this uses his lovely ๐ฑ๐ฏ-๐๐ผ๐ฐ๐ฐ๐ฒ๐ฟ package under the hood.
And a live demo on Streamlit Community Cloud:
I built ๐๐๐ฟ๐ฒ๐ฎ๐บ๐น๐ถ๐-๐๐ผ๐ฐ๐ฐ๐ฒ๐ฟ, a custom React component for Streamlit for working with soccer tracking data.
And a blog post on ๐๐ถ๐ด ๐๐ฎ๐๐ฎ, ๐ง๐ถ๐ป๐ ๐ง๐ฒ๐ฎ๐บ๐
First search term of the 2025 (and my parental leave) was โbaby ate cat foodโ, so weโre off to a good start.
The theory here is that some kind of basic property of โgoal probabilityโ is โoption-nessโ
open.substack.com/pub/absolute...
Really enjoyed the idea of using the sideline to apply equal-and-opposite pressure relative to the player velocity.
Super clever!
๐โโ๏ธโช ๐โโ๏ธ ๐๐ง ๐๐ง๐ญ๐ฎ๐ข๐ญ๐ข๐ฏ๐ ๐๐๐๐ฌ๐ฎ๐ซ๐ ๐๐จ๐ซ ๐๐ซ๐๐ฌ๐ฌ๐ข๐ง๐ ๐ข๐ง ๐๐จ๐๐๐๐ซ
It's been over a year now since I was invited to present my ๐๐ซ๐๐ฌ๐ฌ๐ข๐ง๐ ๐๐ง๐ญ๐๐ง๐ฌ๐ข๐ญ๐ฒ metric to a small group of people at a SkillCorner event in Dallas, TX.
I've finally put some effort into writing a blog on it!
unravelsports.github.io/2024/12/12/p...
We're looking for a ๐๐๐น๐น ๐ฆ๐๐ฎ๐ฐ๐ธ ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐ฒ๐ฟ on a 12-month contract help us build team performance tools across multiple sports.
I think you can get away with relative features like angle and distance to the team centroid combined with a few other heuristics like the number of teammates ahead/behind/left/right.
Similar principals.
