Thank you for all of the work you and Friends of Tracking are doing! I'm new to soccer analytics, but I'm not new to analytics, I'm coming into it with a background in ergonomics, epidemiology and physical therapy. It is fascinating to read the comments and the overall debate. They are philosophically identical to the debate between "evidence based practice" and "clinician expertise" for guiding decisions in physical therapy (and health care in general). And that goes back millennia and the difference between the "universal" and the "particular." Every event, shot and goal has a story and that story is far more nuanced than any model can possibly account for. The question is - as David has very nicely pointed out - how much of the probability of a goal can be understood (universally) from some of the factors that lead to the nuance in any particular situation. It is justifiable to use particulars to build universals, and to use universals to understand particulars. They are both reasonable activities to better understand the game and to extract knowledge for making decisions and interpretations.
Thank you, very interesting overview of xG! I think that what is mostly overlooked when using a human created "Big Chance" feature, is that it represent a huge data drift - the human that tags the shot as a "Big Chance" is aware as far as I know whether the shot ended up as a goal (or at the very least the initial trajectory of the ball). It's a bit of an exaggeration of course , but it would be like using the "Goal" feature to predict and have a perfect model. Therefor looking at the loss function is not enough to indicate that this feature is useful. What might be interesting however that would eliminate this data drift is doing the following: Taking each team's xG on the first half of the season according to a model without "Big Chance" feature, then see the correlation for that team's actual goals (ground truth, not xG) on the second half of the season. Then do the same for a model with "Big Chance" feature, and see whether the results are better. Both model should be trained without data from the second half of that season of course... I would really appreciate your input on this.
I guess Im randomly asking but does anybody know of a trick to get back into an Instagram account?? I was dumb forgot my password. I would love any assistance you can offer me!
@Ignacio Grant i really appreciate your reply. I got to the site thru google and I'm trying it out atm. I see it takes a while so I will get back to you later when my account password hopefully is recovered.
the model u use consider (opponents defensive line patterns) ? Like 1 defender present in box while taking a shot in box and 5 defenders present in box while taking a shot in box, also their shape too.
What about taking into account the direction and speed of the players blocking the shot cone? If a player had a defender running towards them (or not), perhaps that would significantly affect the shot chance.
... "the clubs that use it are some of the most successful clubs".... is there any evidence that using xG helps a team become successful? Or is it that successful teams have the budget to hire the analytics staff to allow them to do xG modeling?
There are two things that I think the expected model has to take into account. 1. Who shot the ball? It is not the same if Messi or Ronaldo shot than other striker, 2. The defenders in front of the shoter, I think that if the shoter has one or more defenders in front of him while shooting it is going to be more difficult to score.
I thought about the same #1 point, but then I reflected: an xG model is the first piece to mathematically model other things, like pass impact and pitch control. So I assume it should be quite generalized, and not be aware of the skills of the particular player, because I want to understand if a situation is GENERALLY a good change for scoring. But I'm not sure about it, what do you think?
Way too many variables to model this. Surely easier to take makret prices into bands for leagues and look at what price bands generate the most expected outcomes, in other words most profitable!
Expected goals is more of a fallacy. It is one of those things that sound so nice and revolutionary but is not realistic. So who calculates the expected goals? Is it the fans? Every game should have an officially released expected goal; we should start taking expecting goals seriously only when that happens. The expected goal will be next to the ball possession, cards, corners and offsides stats in every game
Thank you for all of the work you and Friends of Tracking are doing! I'm new to soccer analytics, but I'm not new to analytics, I'm coming into it with a background in ergonomics, epidemiology and physical therapy. It is fascinating to read the comments and the overall debate. They are philosophically identical to the debate between "evidence based practice" and "clinician expertise" for guiding decisions in physical therapy (and health care in general). And that goes back millennia and the difference between the "universal" and the "particular." Every event, shot and goal has a story and that story is far more nuanced than any model can possibly account for. The question is - as David has very nicely pointed out - how much of the probability of a goal can be understood (universally) from some of the factors that lead to the nuance in any particular situation. It is justifiable to use particulars to build universals, and to use universals to understand particulars. They are both reasonable activities to better understand the game and to extract knowledge for making decisions and interpretations.
Thank you, very interesting overview of xG!
I think that what is mostly overlooked when using a human created "Big Chance" feature, is that it represent a huge data drift - the human that tags the shot as a "Big Chance" is aware as far as I know whether the shot ended up as a goal (or at the very least the initial trajectory of the ball). It's a bit of an exaggeration of course , but it would be like using the "Goal" feature to predict and have a perfect model.
Therefor looking at the loss function is not enough to indicate that this feature is useful.
What might be interesting however that would eliminate this data drift is doing the following:
Taking each team's xG on the first half of the season according to a model without "Big Chance" feature, then see the correlation for that team's actual goals (ground truth, not xG) on the second half of the season.
Then do the same for a model with "Big Chance" feature, and see whether the results are better.
Both model should be trained without data from the second half of that season of course...
I would really appreciate your input on this.
I guess Im randomly asking but does anybody know of a trick to get back into an Instagram account??
I was dumb forgot my password. I would love any assistance you can offer me!
@Grady Gavin Instablaster =)
@Ignacio Grant i really appreciate your reply. I got to the site thru google and I'm trying it out atm.
I see it takes a while so I will get back to you later when my account password hopefully is recovered.
@Ignacio Grant It did the trick and I now got access to my account again. I am so happy:D
Thank you so much you saved my account :D
@Grady Gavin you are welcome xD
thnks 4 taking the time to do ths video
awesome ! Ordered the book - trying to see how I can adapt to help improve field hockey Goal keepers decision making!
the model u use consider (opponents defensive line patterns) ? Like 1 defender present in box while taking a shot in box and 5 defenders present in box while taking a shot in box, also their shape too.
What about taking into account the direction and speed of the players blocking the shot cone? If a player had a defender running towards them (or not), perhaps that would significantly affect the shot chance.
... "the clubs that use it are some of the most successful clubs".... is there any evidence that using xG helps a team become successful? Or is it that successful teams have the budget to hire the analytics staff to allow them to do xG modeling?
There are two things that I think the expected model has to take into account. 1. Who shot the ball? It is not the same if Messi or Ronaldo shot than other striker, 2. The defenders in front of the shoter, I think that if the shoter has one or more defenders in front of him while shooting it is going to be more difficult to score.
I thought about the same #1 point, but then I reflected: an xG model is the first piece to mathematically model other things, like pass impact and pitch control. So I assume it should be quite generalized, and not be aware of the skills of the particular player, because I want to understand if a situation is GENERALLY a good change for scoring. But I'm not sure about it, what do you think?
wht r your thoughts on infogol?
Way too many variables to model this. Surely easier to take makret prices into bands for leagues and look at what price bands generate the most expected outcomes, in other words most profitable!
Expected goals is more of a fallacy. It is one of those things that sound so nice and revolutionary but is not realistic. So who calculates the expected goals? Is it the fans? Every game should have an officially released expected goal; we should start taking expecting goals seriously only when that happens. The expected goal will be next to the ball possession, cards, corners and offsides stats in every game
did u watch the video?