Joined December 2014
StatSheetStuffer.com retweeted
Building WAR for Football #3: Data collection, cleaning, manipulation & clarification ⚽️📊💻 After criticising my own V1 model, in this article I focus on data collection, cleaning, manipulation, and clarification. Including a 12 leagues network. medium.com/@marclamberts/bui…
4
21
2,562
StatSheetStuffer.com retweeted
Alan Nathan has already given us a physics-based calculator for distance given speed+angle. Pretty snazzy excel download Or, if you can stomach small sample sizes, you can rely on actual data baseballsavant.mlb.com/statc…
2
1
1
17
4,261
StatSheetStuffer.com retweeted
Aging curves for umpires!
Replying to @903124S
Using delta method by @tangotiger, given most umpire start in their late 30s call/strike accuracy strictly decrease with age and decrease by ~0.75% from age 30 to age 50 and accelerate with decrease of ~0.75% from age 50 to 65 2/n
3
6
161
24,805
StatSheetStuffer.com retweeted
This article is another excerpt from the paper and will focus on case studies of other players, specifically how xPTS-RAPM can quantify the extent and value of a player's impact on shot diet. xPTS-RAPM and its sister metric, Conversion RAPM (cRAPM), also provide a way to decompose RAPM away from an opaque numbers
1
2
19
1,440
StatSheetStuffer.com retweeted
Using delta method by @tangotiger, given most umpire start in their late 30s call/strike accuracy strictly decrease with age and decrease by ~0.75% from age 30 to age 50 and accelerate with decrease of ~0.75% from age 50 to 65 2/n
6
17
1
89
34,432
StatSheetStuffer.com retweeted
Building WAR for Football #1: My first attempt at Wins Above Replacement ⚽️📊 A concept of WAR with many flaws, but sharing the whole process is important. Only portraying successful version isn't reality. Methodology, calculations and results ⤵️ medium.com/@marclamberts/bui…
5
2
51
13,690
StatSheetStuffer.com retweeted
a free computer vision library crossed 50,000 github stars this week, and the maintainer's post said thank you instead of announcing a pricing page supervision, by Piotr Skalski at Roboflow. MIT licence, 5,131 commits, 1,732 pull requests, 188 contributors, started in november 2022 what it actually does, because the name says nothing: → you run any detection model, YOLO or anything else, and supervision handles everything after the model → draw the boxes and labels the way you want → track the same object across frames so it keeps its number → draw a line and count what crosses it. draw a zone and count what's inside → the basketball and traffic footage below is that, on real video why this is the kind of project worth caring about: for years the hard part of computer vision was the model. now the model is a download, and the hard part is the plumbing around it: the tracking, the counting, the annotation, the dataset conversions. that plumbing is what got written 5,131 times here and given away → pip install supervision → github.com/roboflow/supervis… fifty thousand people starred it. the demo takes about 20 lines footage is from @skalskip92's own post
1
16
1
184
18,492
StatSheetStuffer.com retweeted
First time on the @DrivelineBB YT in nearly 2 years here! @connerpelletier and I present our Saberseminar bat path biomech analysis we made to help our in-gym and remote athletes. Watch ⬇️ youtube.com/watch?v=7j00kQje…
9
31
6,741
StatSheetStuffer.com retweeted
𝐇𝐨𝐰 𝐭𝐨 𝐁𝐮𝐢𝐥𝐝 𝐚𝐧 𝐄𝐟𝐟𝐞𝐜𝐭𝐢𝐯𝐞 𝐏𝐥𝐚𝐲𝐞𝐫 𝐈𝐧𝐝𝐞𝐱 🎯 Creating a role-based player index is one of the foundational stages of a scouting process. It is also one that requires a special level of scrutiny: a flawed approach can lead to a strong candidate being filtered out too early or an unsuitable one making it through to the next stage. Getting it right requires a mix of football knowledge and a basic understanding of the underlying mathematics. In our latest blog, we break down how to build an effective player index, including the three key risks to watch out for. Check it out 👇 skillcorner.com/articles/how…
1
15
4
174
61,733
StatSheetStuffer.com retweeted
Introducing NFL SNAP SNAP stands for “Seven Numbers Assessing Performance” Inspired by the “How badly did we get beat” chart, I wanted to create a broader picture of why a team won/lost. SNAP includes: -Success Rate -Run/Pass value difference -Explosive play net -Turnover value net -Special Teams value net -Penalty value net
33
58
16
844
236,962
StatSheetStuffer.com retweeted
🚨 NEW @drawtactics feature Introducing Player's Eye Ever watched a game and wished you could look at the scene from a player's perspective? drawtactics.com now allows exactly that Simply upload a screenshot and then choose your player (I'll work on making the 3D models look better lol)
2
27
3
214
18,349
StatSheetStuffer.com retweeted
At @SkillCorner , we just launched our repo with basketball open data! github.com/SkillCorner/opend… You'll find tracking, event, and aggregate data samples from ACB's 25/26 season. This is an unprecedented opportunity to dig into an international league's tracking data.
3
17
6
169
47,235
StatSheetStuffer.com retweeted
Pitch Profiler has open sourced a reconstruction of Baseball Savant's swing timing data! Savant publishes where on the bat, timing, and height above the swing plane as season aggregates. This rebuilds all three per swing, cutting the error against Savant's pitcher rates 57%.
1
4
51
8,337
StatSheetStuffer.com retweeted
Driveline Baseball x Roboflow Come see @SamEhrlich and I speak in San Francisco at the @roboflow Visual Intelligence Summit on October 22nd. @DrivelineBB is pushing the boundaries of applied AI/CV in sports and we're honored to be selected for VIS! summit.roboflow.com/
1
5
1
41
12,301
StatSheetStuffer.com retweeted
It looks like there is a relationship between how big your breaking ball is and how well you can command it. Command here provided by @open_command, pitch numbers by @FanGraphs (darker is better command, aka smaller miss distance from target)
8
6
1
91
22,540
StatSheetStuffer.com retweeted
I built a simulator that re-prices luck in an NFL game. Every loose ball, field goal, and dropped pass is rerun 160K times. Introducing the NFL Deserve-to-Win % Built on data gathered via #nflverse All code, models, and research can be found here: github.com/dgrifka/nfl_simul… 🧵
15
28
8
482
201,618
StatSheetStuffer.com retweeted
I have uploaded all NHL play-by-play data in a per-season format in the form of a Kaggle dataset, if anyone is interested. kaggle.com/datasets/owensing…
2
4
42
2,552
StatSheetStuffer.com retweeted
Stop treating every completed pass as equally valuable. A lot of passing analysis still rewards volume: - more completed passes = more influence -more connections = greater importance Useful, but incomplete. 10 safe CB-to-CB passes can dominate a conventional passing network. 3 line-breaking passes that move the ball into dangerous areas may matter far more. A 2026 paper tackles this with a Valued Passing Network (VPN). Instead of weighting connections only by frequency, it weights them by the Expected Threat (xT) they create. So the question changes from: “Who passed to whom the most?” to: “Which passing relationships actually generated threat?” Using the 2022 World Cup, the authors compared conventional passing networks with xT-weighted ones. The valued network showed stronger relationships with player ratings, particularly for midfielders and forwards. The figures make the distinction clear. A conventional network is good at showing structure and circulation. A valued network highlights which connections actually increase attacking threat. That helps separate: - volume from impact - involvement from contribution - circulation from progression - safe possession from threat creation And the World Cup final example is especially interesting: some players who look central in the conventional network become less prominent once passing value is considered, while players such as Messi, Mbappé and Griezmann gain relative importance. This doesn’t mean safe or backwards passes are bad. They can be tactically essential. It means that a completed pass is not automatically a valuable pass. For scouting, that leads to better questions: - Which midfielders generate value rather than simply accumulate touches? - Which player combinations consistently move possession into dangerous states? - Who looks average by passing volume but becomes important when contribution is considered? Pass volume ≠ pass value. A player can be central to circulation without being central to threat creation. Ideally, we should measure both.
2
14
1
109
7,314
StatSheetStuffer.com retweeted
Slight work in progress still, but hoop-explorer now has an "Exports" page describing the new home for the batch data I generate. If you're in my "fetch data from dropbox every morning" club, you should switch to here for the upcoming season :)
1
2
21
1,026