David Sumpter
@Soccermatics
Followers
33K
Following
10K
Media
2K
Statuses
10K
Professor of Applied Maths and Author. Co-founder and data scientist at Twelve football.
Uppsala, Sweden
Joined January 2015
Wordalisations: automated explanations of data. It is my great pleasure to announce that @twelve_football have created an open-source educational version of our Twelve GPT product. Now you can learn how to create football and other wordalisations!.
1
8
48
Do you want to learn football analytics? 🤔. You have come to the right place! ✅. My course 'Soccermatics: mathematical modelling of football' is available now. AND IT IS FREE.🤗.
69
527
3K
Look in the bottom right corner of this graph. 👀. The two teams with the lowest xG/shot against and highest xG/shot for are owned by professional gamblers who made their fortunes using expected goals. Football analytics really does work.
26
194
2K
Expected Threat (xT) is one of the most important (but least well-understood) statistics in football. ⚽️⚽️. It was used by Liverpool (they call it Goals Added) to do some of their best scouting. 📈. And now, YOU can learn all about it. 🧵.
22
203
2K
Arsenal are very good this season. Here is a thread with the numbers which explain why. (Yes, of course being 1st in the league is also an important number😉). Before today they have outperformed their opponents on xG in every match.
11
207
1K
A thread on Liverpool using numbers. Six games is great for using expected goals and expected threat to find out what is going on. First of all they have 'won' all six on xG.
49
196
1K
Hammarby are doing something of a first in football analytics. The club has set up metrics for their style of play, which will be communicated throughout the season with fans, the board and players alike.
14
79
907
The complete guide course to football analytics is now available for free. Covering everything from getting started in python, through expected goals, simulating matches and evaluating players, to tracking data and expected possession value. Enjoy!
6
84
584
The Future of Life Institute is a problem. Being in same age-group (lower end maybe😃)/cultural background (8-bit programming📼⌨️) as these men, makes me feel uncomfortable and embarrassed for them. Here they (Musk, Hassabis, Tegmark, Bostrom etc.) are in 2017.
14
112
458
How should we do mathematical modelling? . With more and more strange ideas flying about AI (It can read minds! it is truly intelligent and might take over the world!) we need to refocus. This is what I do in this paper with Linnea Gyllingberg and @Abebab.
7
94
452
You might not be interested in football or maths, but you HAVE to watch @maramperninety here. A 20-year-old Muslim woman, she is one of the absolute best at combining these two (otherwise) male-dominated areas. She talks so eloquently about her journey.
5
80
455
Mathematical modelling in football course starts 31st of August. Free to all EU citizens. Full syllabus and details how to register here:
31
75
455
If we are going to be able to scout players using data, we have to use an understanding of the game. I explain how to do this using Professor Paco Seirul·lo’s zonal method.
3
54
425
It never ceases to amaze me how close goal distributions are to Poisson. This is distribution of number of goals in a match since 2005 in the Premier League. Bars are histogram. Line is Poisson distribution. 0-0 draws are the only outlier.
15
29
394
News from me. I am now working as data scientist at @Hammarbyfotboll football club. I will be active at all levels of the club: from working with players to communicating with fans. Bringing top quality analytics to #bajen.
46
19
394
The value of disruptive runs. Or. why Firmino is such a valuable player for Liverpool. New work by @jernejfl using tracking data to evaluate off-the-ball movement. @SkillCorner @twelve_football
11
69
377
They are best in league at getting the ball in to the final third and in to the box per match.
5
41
359
Here is my list, based on some research I made during my course, who should get credit for first application of things in football analytics. Expected goals, Expected Threat, Pitch Control etc. .
5
52
351
The Masterclass is Ready! Mathematical Modelling Course in Football videos now in a start to finish playlist. With amazing lectures by @EightyFivePoint, @the_spearman, @suds_g, @lucpappalard, @JanVanHaaren and @petermckeever.
6
49
321
The first video lecture in 'Soccermatics: mathematical modelling of football' is here! 🤗. Learn how to make fantastic pitch visualisations. 💡.
4
25
291
The genius of Barcelona players. Xavi takes in information very fast. Iniesta creates unique solutions using the info. Dotted line is population average.
5
71
273
An innovation in latest release of @twelve_football TwelveGPT is a transfer model. Here is how players perform as they move between leagues in terms of passing quality. Players coming from La Liga can hold the same level of passing after a move to the Premier League. But.
8
33
291
Want to learn how to build a neural network model that uses @statsbomb 360 data to create an expected goals model which accounts for player positions when shot is taken? . @aleksander_and has created a step by step implementation of exactly that!.
3
22
275
Applications now open for Mathematical Modelling of Football. This course is taught by me in Sep-Oct for 8 weeks (10-15 hours work per week). Deadline for application is 15th April. Application here: Course website 2021:
10
70
272
When a football player peaks in performance depends very strongly on what you want from them. Model fit on event data from seven seasons of Premier League, La Liga, Bundesliga, Ligue 1 and Series A.
2
29
268
It takes attention from the real research that has been done in this area, by Gebru, Mitchell, Bender, O'Neil and many others who have worked on good data practices, the costs associated with AI and the real dangers involved.
3
55
234
We (myself and @KozlovaNicole at @twelve_football ) have modeled how players perform in different metrics when they move between leagues. Here is the overall ranking of leagues, looking at how (on average) metrics change when players transfer. Red is worse, green better.
13
32
259
When scouting players we want measurements to be repeatable over seasons: that a player can do what they did previous season in the coming season. Finishing does not have that property, even over a season. Each dot below compares the finishing of a player between two seasons.
13
42
254
Great project. Look forward to seeing more from this.
almost fully functional version of my football AI project. today, I added player tracking using ByteTrack and projection of players onto the map. code coming soon:
2
18
255
Det finns en klubb som jag jobbar med som kommer alltid ligger närmast min 💚. Tack att jag fick vara med ikväll, träffa så många av er och prata ”nyckeltal”
5
0
249
Arsenal just sold their 2nd best player to Liverpool. More defensive impact that Xhaka, twice the xG of Giroud and passing that rivals Özil!
28
161
227
Their most dangerous combination is Ben White to Saka, closely followed by Odegaard to Saka. Jesus is receiving the ball well from everyone. Plot shows expected threat (value added by passes) between pairs of players.
2
27
228
Lesson 2 of 'Soccermatics: the course' is live! ⚽️. ✅Learn how to make an expected goals model. ✅Does possession influence match outcome?.✅Test whether your favourite football myth is true. Get started with this video:.
5
23
235
How Swedes were fooled by one of the biggest scientific bluffs of our time. Think that people can be classified as colours? Think again. Over New Year I translated and edited Dan Katz investigation in the 'Surrounded by Idiots'.
14
78
220
This is exactly why we set up #FoT. Liverpool FC's William Spearman (@the_spearman) gives a masterclass in pitch control and how tracking data can be used on the field of play.
4
34
225
The Soccermatics course:. ✅All teaching materials free forever.✅Covers all coding, machine learning, maths and data science skills.✅Given in Student, Professional and Enthusiast versions.✅ Includes graded individual & project work. Take a look:.
3
26
228
And Odegaard is alsovery important. He is involved in pass chains that create an average of 1.12xG per match! . These are his passes with an xT value of over 0.05. Most of them directly forward in to the box.
2
30
214
What is the best way to learn football analytics?. Out on the pitch, of course!. I have just come back from a course with the Polish federation for analysts, where we started exactly there.
2
17
223
The first question for football teams wanting to use analytics is choosing what to measure.
1
31
211
It is live! How to build an expected goals model. First part of my step by step guide using @wyscout freely available football event data. Everything you need to understand and build your own model in Python.
4
27
206
As I always say . you have to wait at least five matches to trust expected goals to give a picture of how good a team really is.
7
14
201
Their strength is in converting final third entries in to box entries (this is where Jesus is key).
1
13
189
Mathematical modelling of football course is up and running again. All the material (lectures, tutorials etc.) is available here: You can't register now but you can follow the progress. And you are welcome next year!.
2
29
192
Should you write about real goals or expected goals? A guide for journalists.
5
55
188
Arsenal deserve to be where they are in the league.(xG before todays match). It is going to be an exciting season for the Gunners!
3
15
185
I have written a book about Ten Equations that can change your life for the better. Out 1st of October with @AllenLaneBooks, but you can get a head start here: What Are The Ten Equations?
8
35
186
Hammarby dominated the match from start to finish. 2.55 against 0.75 in expected goals. Worthy winners of the Stockholm Derby.
1
35
186
Dear Jurgen Klopp @LFC. I have been watching your team play and compared it to some value models. It is straightforward. Balls out wide to TAA (by Thiago for example, shown below) do have value. (1/n)
14
36
174
Advice for anyone who wants to become a football data scientist. All the most important information from last night's broadcast. via @YouTube.
6
25
170
Is the model wrong or is it everyone else?. Phil Foden is obviously a great player. But TwelveGPT puts a lot of this season's excitement about him down to some good finishing.
13
14
172
Very proud to announce my first article for @TheAthletic. A look at how to beat the odds in the World Cup.
4
8
172
If the league was decided by simulating it 10,000 times using xG then today Manchester City would have won by 8.9 points while sending Brighton down. Wolves would have qualified for the Champions League at Anfield. Meanwhile, Cardiff just secured survival against Man United.
10
42
168
"I've never been in a team that dominated a match like this", said Tankovic afterwards, "we rolled over them." . Top down look at how Hammarby's press produces goals. Data from @signality. Pitch control by @JaviOnData. Work as part of Fran Peralta's masters thesis.
6
24
169
It is on! If you would like to learn football analytics and get University credits then this is the course for you. FREE TO EU/EEA CITIZENS. #FoT
14
30
163
Of course, the league position and the results are important statistics too!. But overall, Liverpool are (together with Arsenal and Manchester City) one of the strongest three teams in terms of underlying numbers. (Figure not updated for today's results)
7
17
163
Today was no exception. Spurs total includes a 0.75xG penalty!
2
12
163
After seven games played there is one team that stands out in the quality and quantity of chances created in Allsvenskan. The darker the shading the higher the average xG. The darkest green in #bajen.
6
21
168
Want something to do to take your mind off things? I have put up all my lectures for Statistical Machine Learning. The course is built by the excellent machine learning group in Uppsala. Google Colab notebooks and pen and paper exercises throughout.
2
21
164
Lesson 3 of Soccermatics: mathematical modelling of football is available now! Learn how to make player radars and rank players. What to include, how to correct for possession and to get the right context.
2
24
159
This evening the Friends of Tracking Youtube channel will be announcing an exciting new collaboration with @SkillCorner. Free tracking data of the best teams in the world, freely available to all! Watch this space #FoT.
7
15
163
And those box entries result in shots. (The Bournemouth result influences these results somewhat but the numbers still look good).
2
20
153
New for me for 2024. This year I will work half-time for @twelve_football. I have always had Twelve as a 'hobby', but now interest amongst clubs has really taken off.🚀. We will build the most comprehensive tool for data analysis - TwelveGPT: football in words and visuals.🤖⚽️
10
2
155
But the single most dangerous pairing is TAA to Salah. This matrix shows expected threat of passes between players.
1
8
154
Friends of Tracking is back! Creating a pass flow graph in R by @MishraAbhiA. Really clear presentation.
4
19
156
I just published “How an algorithm can measure defence and press in football.”
3
44
154
I have the best job in the world, talking space creation with the best attacking team in Sweden. And better than that I get to tag @leobengtsson10. Good discussion. Thanks boys.
2
2
152
The most exciting league in the world and this morning we are top of it. Feels very good to go in to work today. What a team!
3
7
155
Using experted goals with players. Here is a video i made with @axelpershagen explaining how we work at Hammarby. In english from 1:00.
6
21
149
Another individual skill radar. Saka is effective because he does things quickly. He ranks highly in making rapid passes (top 15%), producing value in xT (top 10%) and dribbling at high speed (top 20%). Most of all, he can pass the ball effectively when play is moving fast.
5
21
141
The next Friends of Tracking Youtube live talk on Thursday will look at "Open resources for getting started in football data science". I'll start a list in the thread below. Feel free to add to help me prepare the talk.
1
33
148
Little of what football commentators say seems motivated by facts. One narrative is Salah is playing wider this season. Below is xT for 3 PL games, compared to same opponents last season. Make up your own mind🤔 . It would be so easy to show things like this on TV.🤩
4
14
145
How Bojanic, Tankovic, Kacaniklic and Djurdjic open up defences. No wonder we caught the interest of Ibrahimovic. Green are areas where pass success probability is higher, red are less probable.
2
9
146
Kacaniklic goal against Elfsborg. Amazing team-buildup and control of opponent's space. Especially Tankovic (22) first involvement and then run for the assist. The finish was all Alex, though!
5
9
150
Sometimes the numbers just make sense!.
8
5
147
Want to learn how the maths and coding needed to build a betting model for the World Cup?. I have got you covered!. I have created a step by step guide to The Betting Equation from my book The Ten Equations That Rule The World.
3
18
146
Is Pedri equal to Kante plus Jorginho? . A bit of a cheeky title, but it is interesting to look at how tracking data can be used to show why Pedri was so good in La Liga last season.
6
32
136
Explaining Expected Threat.. Nice to see everyone getting in to a method we have been using at @twelve_football and made publicly available for the last 4 years 😎. But even more important to remember that it came from @srudd_ok over a decade ago 😱😎😎
6
22
136
The Human Histogram. Demonstrating probability distributions using the power of crowds.
0
108
134
Richard Magyar. A hero in so many ways and also the statistically best player today.
0
11
141
How do you turn your football analytics in to a beautiful app?. I explain it all step-by-step in this video: Thanks to @streamlit for your amazing product and @jernejfl at @twelve_football for creating beautiful visuals.
3
9
138
How should we measure the performance of a machine learning model? 😕. When I was teaching ML for the first time last year, I was surprised to find there was no agreed upon single number which measures model performance. 🤯. So I decided to look at the question myself. 🧵.
5
19
132
I was so busy after finishing the Soccermatics free football analytics course I never got round to sharing all the different parts we created. I think I stopped at expected threat, but I will recap. Here is the concept explained:
3
6
141
I just talked to an AI bot about football. How cool is that! 🤖🤖🤖
10
26
131
Pitch control is one of the most important ways of understanding movement of football players. In this video I give an overview of the concept.
2
21
138
I never dreamed that one day I'd make it in to 'Roy of the Rovers', but life takes some strange turns.
5
9
132
Paulinho was per minute one of the absolute best players in Allsvenskan last season. Here is why. (1/5)
3
18
130
The lectures for the 2nd week of the Modelling Football course are up now. We develop expected goals and other statistical models of football.
1
19
132
Yesterday really was freedom day for me. I got my second dose of vaccine. But most importantly I got this. Not a freedom I asked for, but one I am happy to receive.
5
0
131
I made this as an 'exercise' for my course. Wanted to think about how KPIs can help at this stage in the season to give some perspective.
13
5
123
Turns out that @theresa_may was only slightly more influential than @LordBuckethead during election Twitter.
3
34
114
Where Chelsea regained possession (line thickness=num actions) @ChelseaStats. Oscar showing Kante how its done!
8
153
125
Using analytics to create patterns of play for @BarcaInnoHub. I look at how Manchester City implement the "5-second-rule" using pitch control. It is all about where you are before you lose the ball.
4
15
125