mark thompson
@markposts.bsky.social
4.2K followers 590 following 2.7K posts
data software engineer @ Twenty3 Sports || sport for all || (he/him) || ⚽🎾💻📰🏎️🤖
Posts Media Videos Starter Packs
markposts.bsky.social
Oh, I was thinking more like how differentiating 'wing play teams' kinda made sense in the Premier League (at one point anyway) bc a lot of teams tried to play centrally, but it probably flattens different wing-based approaches that in e.g. WSL might be more meaningful to separate
markposts.bsky.social
Y'know the saying about "a hundred words for snow"? Language develops to be specific about things that are common and meaningfully different (Britain has lots of references for rain). Different ecosystems of football have slightly different places where differentiation is meaningful
markposts.bsky.social
To take a different approach to this to my other answer, I think 'simple, explainable ways of describing teams and players' has a lot of life left to give.

*Particularly* if you're interested in a league outside the 'Big 5' and wanna consider that as a similar-but-different ecosystem
catabush.com
What is an already existing metric/area of study that you think hasn’t been given enough attention (gamestate, phase of play, etc.)
markposts.bsky.social
Similarly, stuff around xG/post-shot xG and the margin for error on players' shots. That's had a little more work done on it publicly (will link some later if I remember) but not much considering how fundamental it is
markposts.bsky.social
Because, when doing that kind of modelling, you have to account for the fact that e.g. 60 yard passes incomplete in the opposition box might just be an attempted 30 yard pass that missed and ran through to the goalkeeper
markposts.bsky.social
Oh shoot. This is a good question.

I think pass intentionality is something that is an intrinsic part of the sport that receives very little attention. I've only ever really seen it discussed when people are talking about xPass models
catabush.com
What is an already existing metric/area of study that you think hasn’t been given enough attention (gamestate, phase of play, etc.)
markposts.bsky.social
This is turning out fun, as hoped, and a nice way of getting through ironing, keep 'em coming
markposts.bsky.social
Football data/analytics/stats Q&A?

['About Me' for newcomers: I've worked in the biz for 6 years now, and was poking around spreadsheets before that, and I want to justify procrastinating the ironing]
markposts.bsky.social
(Related, I think too much chatter over United men signing Mbeumo and Cunha goes like 'they bought xG overperformers' as if it is thoroughly reported that they bought them for that, when those players also fit Amorim's system to a T. The xG thing is notable but explain *why*)
markposts.bsky.social
It's not like "phwoar, look at that overperformance, what a finisher", which is good, but it's kinda like "look at the amount scored more than xG - that's, y'know, kinda good".

It's useful (and longer-term there might be signal), but it's like, picking up the best avocado is useful & nice
markposts.bsky.social
I feel like I should have a good answer for this but don't on a mass scale, so I'll do a smaller-scale answer

Public conversation around goals vs xG is really weird.
christina6ys.bsky.social
What do you think is the most misused stat/incorrectly applied measure in football currently?
markposts.bsky.social
Your early projects will have a lot of time just getting to grips with data - and with coding if you're new to coding - so little 'trinket' studies can be great for keeping your momentum up
markposts.bsky.social
In general, my opinion is that your early projects are better if they're "poke a bush with a stick" projects. Small, not trying to 'solve' something, probably won't have a clear outcome, but based around a personal curiosity you have. E.g. which GK has the biggest % of passes outside their box?
markposts.bsky.social
If coding seems daunting and you just want to look at a bunch of data, you can copy and paste big tables off FBref into Excel pretty easily. They have a 'combined Big 5 Leagues' table. Practice making some of your own metrics, 'passes per foul', things like that
markposts.bsky.social
If coding seems daunting and you just want to look at a bunch of data, you can copy and paste big tables off FBref into Excel pretty easily. They have a 'combined Big 5 Leagues' table. Practice making some of your own metrics, 'passes per foul', things like that
markposts.bsky.social
It requires coding, but Statsbomb have a Python & R package, and there's probably enough public code that an LLM can help you produce working code for it very easily.

I'd recommend finding something involving pass maps:
- easy to style
- relevant to all players
- easy to test (check a GK & RB)
markposts.bsky.social
Depends a bit on the level of beginner

Statsbomb public dataset is a good starting point for its range: international tournaments, full club seasons, different eras. If you want to make a shot map in a match from every decade since the 1970s, you can do that

github.com/statsbomb/op...
markposts.bsky.social
I feel like there's already a provider somewhere who I'm forgetting with stats like "passes received with open stance between the lines". If there's not, that's the kinda thing you could get. Or the coordination of a defensive line attempting an offside trap.
markposts.bsky.social
One possible future is that it's not a pressing priority for providers, they don't productise it, and random clubs fumble around in isolation to find the most useful features. Another is that a smart provider builds a targeted set of features and everyone copies their innovation
markposts.bsky.social
The data? Hugely, although most of it is already in place, probably being produced *somewhere*. Pose data could isolate shooting technique and basic (but important) body orientation

The extent that that is used depends heavily on data companies though
oli-econ.bsky.social
How much do you think football data will change in 10 years
markposts.bsky.social
No question too small, *some* questions too big
markposts.bsky.social
Football data/analytics/stats Q&A?

['About Me' for newcomers: I've worked in the biz for 6 years now, and was poking around spreadsheets before that, and I want to justify procrastinating the ironing]
markposts.bsky.social
Football data/analytics/stats Q&A?

['About Me' for newcomers: I've worked in the biz for 6 years now, and was poking around spreadsheets before that, and I want to justify procrastinating the ironing]
markposts.bsky.social
Article says B2 instead of B1 fwiw
markposts.bsky.social
And my personal one-to-try-and-avoid, although it's a tough default to break out of, "think"
timetravel.camp
One possible way out of LLM Psychosis is to enact the discipline of refusing to use words that ascribe agency to LLMs and the outputs thereof.
An LLM does not “suggest”, “explain”, “create”, “plan”, “prioritize”, etc.
markposts.bsky.social
I find it kinda funny that there are companies incentivising token use, but it also seems that a full/large context window can lead to worse model performance
Reposted by mark thompson
pysport.org
We’re excited to launch the 𝐒𝐤𝐢𝐥𝐥𝐂𝐨𝐫𝐧𝐞𝐫 𝐗 𝐏𝐲𝐒𝐩𝐨𝐫𝐭 𝐀𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐬 𝐂𝐮𝐩 - a hackathon-style challenge that will bring together the community to collaborate, compete, and innovate using open-source SkillCorner data.

👉 Sign up pysport.org/analytics-cup
markposts.bsky.social
Also there should be book recycling boxes at supermarkets