Im worried 😟

[NoloBoro]

Too many cry babies posting in here. Early days yet. Unfortunately, if you gave the average Teessider a bag of gold, they’d whinge about how heavy it was!!

 

[Super Bernie]

Yes early days, start to worry after 15 games. Keep the faith

 

[Laughing]

It IS quantifiable just not to a 100% scientific standard.

Cool quantify it for me please, because I can't. I looked at this and ran some ml deepnet learning over it and it doesn't work as a predictive tool. My assumption at the time is it's based on watchers categorising chances. The issue here is that 2 different observers will categorise differently.

It really isn't quantifiable, it's too opinion based.

 

[Rofesleg]

We were good in spells against Stoke,

Good in spells? That's how you view our performance against Stoke?

 

[Rofesleg]

Cool quantify it for me please, because I can't. I looked at this and ran some ml deepnet learning over it and it doesn't work as a predictive tool. My assumption at the time is it's based on watchers categorising chances. The issue here is that 2 different observers will categorise differently.

It really isn't quantifiable, it's too opinion based.

I don't think xG and the like were ever intended to be used as predictive tools, more a way of analysing performances to identify strengths and weaknesses for coaching teams to work on.
That is probably why your Skynet AI System can't utilise it the way you would like it to.

 

[Laughing]

I don't think xG and the like were ever intended to be used as predictive tools, more a way of analysing performances to identify strengths and weaknesses for coaching teams to work on.
That is probably why your Skynet AI System can't utilise it the way you would like it to.

Possibly. More likely it isnt consistent enough. I have just over 10k games for training purposes. Over that number of games, it should even out if the statistic was any good. Stats like goals scored, ppg, goals before ht all contribute much more, I suspect because they are definitive and not subjective in any way shape or form.

With a nueral net, it cannot tell you why it reaches a decision, but it can show heat maps to help you underrstand how it reached a decision. xG has such a low impact as to be best ignored, when compared to other statistics, obviously.

 

[BoroMart]

Cool quantify it for me please, because I can't. I looked at this and ran some ml deepnet learning over it and it doesn't work as a predictive tool. My assumption at the time is it's based on watchers categorising chances. The issue here is that 2 different observers will categorise differently.

It really isn't quantifiable, it's too opinion based.

it's about ball position, angle of shot, position of GK, and shot type. Those are facts not opinions, but doesn't take into account the speed of the pass, cross, proximity of defenders, It's not perfect and the video angle can be difficult to get 100% accuracy, but you will get a close proximity and in fact there is no reason why it can't be automated.

I don't really buy xG as a predictive tool, but for performance analysis it's useful and has enough accuracy and consistency to be of worth.

 

[Laughing]

it's about ball position, angle of shot, position of GK, and shot type. Those are facts not opinions, but doesn't take into account the speed of the pass, cross, proximity of defenders, It's not perfect and the video angle can be difficult to get 100% accuracy, but you will get a close proximity and in fact there is no reason why it can't be automated.

I don't really buy xG as a predictive tool, but for performance analysis it's useful and has enough accuracy and consistency to be of worth.

I don't disagree with much of that. However it isn't quantifiable unless 2 independent people watching the same game come up with the same value. They wouldn't it's too opinion based.

It's worthwhile if you haven't seen the game to give an idea about a sides dominance. Not much more than that. Shots on/off target are better predictive stats both in play and to forecast future performance.

 

[Rofesleg]

I don't disagree with much of that. However it isn't quantifiable unless 2 independent people watching the same game come up with the same value. They wouldn't it's too opinion based.

It's worthwhile if you haven't seen the game to give an idea about a sides dominance. Not much more than that. Shots on/off target are better predictive stats both in play and to forecast future performance.

Well if doesn't work, then don't use it to try and predict future performance. I'm not sure what the problem is.

 

[Laughing]

Well if doesn't work, then don't use it to try and predict future performance. I'm not sure what the problem is.

I wasn't aware there was a problem?

 

[Andy_W]

I'm not worried, maybe 5%, but that's only due to morale/ frustration.

We're playing well, have not had a main striker and have been exceptionally unfortunate at the back and with decisions, and not put enough away at the other end.

If we had the full squad, were playing crap, getting battered, had played loads of crap sides (especially at home), had loads of decisions go our way, and still had next to no points, then I'd be worried but none of it is true.

 

[Laughing]

I'm not worried, maybe 5%, but that's only due to morale.

We're playing well, have not had a main striker and have been exceptionally unfortunate at the back and with decisions, and not put enough away at the other end.

If we had the full squad, were playing crap, getting battered, had played loads of crap sides (especially at home), had loads of decisions go our way, and still had next to no points, then I'd be worried but none of it is true.

I would largely agree with this. We probably don't have the number of points we deserve at the moment. The flip side is, if this continued, you would start to wonder if it was bad luck or bad judgement. We are nowhere near this point at the moment.

 

[festa5]

Suppose xG indicates we're missing too many chances at present. But we are creating them.

It also suggests the teams we've played so far have been more clinical. With Sheff United and West Brom that might be the trend with them for the season. They have strong players so might just be better at taking their chances. QPR, Stoke and Reading are unlikely to score better than their xG in most matches, to that extent we're a bit unlucky they've done it against us.

For all that everyone agrees we were poor v Reading, they scored a goal that was incredibly unlikely from a player very unlikely to score it. We created a number of chances better than that one.

Play that game again 10 times, with those respective performances from both teams and we'd probably win it 2/3 times and draw the rest. And that is us playing well below our expected level.

Whether you call it "luck" or whatever it doesn't really matter. But I don't see what in the performances (as opposed to results) is causing the doom and gloom from some posters.

We're doing a lot more right than we're doing wrong. If we continue to do that, the results will come. The concern is that the lack of results starts to have a negative mental impact and that causes the performances to drop off further.

 

[Laughing]

Suppose xG indicates we're missing too many chances at present. But we are creating them.

It also suggests the teams we've played so far have been more clinical. With Sheff United and West Brom that might be the trend with them for the season. They have strong players so might just be better at taking their chances. QPR, Stoke and Reading are unlikely to score better than their xG in most matches, to that extent we're a bit unlucky they've done it against us.

For all that everyone agrees we were poor v Reading, they scored a goal that was incredibly unlikely from a player very unlikely to score it. We created a number of chances better than that one.

Play that game again 10 times, with those respective performances from both teams and we'd probably win it 2/3 times and draw the rest. And that is us playing well below our expected level.

Whether you call it "luck" or whatever it doesn't really matter. But I don't see what in the performances (as opposed to results) is causing the doom and gloom from some posters.

We're doing a lot more right than we're doing wrong. If we continue to do that, the results will come. The concern is that the lack of results starts to have a negative mental impact and that causes the performances to drop off further.

Your last point is particularly valid. It can also cause the manager to doubt the approach though I would hope that Wilder is experrienced enough to know it will work itself out. Remember Monk buying players for a 433 then abandoning the entire approach?

 

[Andy_W]

I don't disagree with much of that. However it isn't quantifiable unless 2 independent people watching the same game come up with the same value. They wouldn't it's too opinion based.

It's worthwhile if you haven't seen the game to give an idea about a sides dominance. Not much more than that. Shots on/off target are better predictive stats both in play and to forecast future performance.

xG is a computer algorithm which has analysed 300,000 shots, from Opta, who are the daddy in stats. It factors in every player's position (with/ without the ball), assist type, header/ good foot/ bad foot, angle, distance, one on one's etc, it's going to be far more accurate than 2 professional analysers or 10 fans watching a game, and maybe similar accuracy to 100 people watching a game, but only because their errors might even out. I don't think it factors in which player is taking the shot, but it shouldn't, as it's not a player measure, it's a measure of chance quality.

It's like on Saturday, I said we were unlucky as that chance they scored probably had a 1 in 50 of going in, it was actually 3%, so 1 in 33, so I was a little off, but I mentioned the guy who scored it hadn't scored for a 100 games, so it wasn't like De Bruyne hitting it. For him to score that was probably a 1 in 100 or more. Then also it should have been given offside as Steffen's view was blocked.

Se below why Mowatt's chance was 10x better, hard to disagree with.



It's 5x more reliable than shots/ shots on target, as it's putting a value on each of them which is probably accurate to 5-10%. 10x 25-yarders isn't worth the same as a chance on the penalty spot etc.

It's like me giving you the option of a Penalty (probably 80% chance) or three 25 yard strikes, nobody is going to pick the 3 x 25 yarders, which makes the shots/ shots on target very unreliable unless they all have similar value, which they don't and there's never enough in one game to make it a measure to rely on for that game.

 

[Andy_W]

Suppose xG indicates we're missing too many chances at present. But we are creating them.

It also suggests the teams we've played so far have been more clinical. With Sheff United and West Brom that might be the trend with them for the season. They have strong players so might just be better at taking their chances. QPR, Stoke and Reading are unlikely to score better than their xG in most matches, to that extent we're a bit unlucky they've done it against us.

For all that everyone agrees we were poor v Reading, they scored a goal that was incredibly unlikely from a player very unlikely to score it. We created a number of chances better than that one.

Play that game again 10 times, with those respective performances from both teams and we'd probably win it 2/3 times and draw the rest. And that is us playing well below our expected level.

Whether you call it "luck" or whatever it doesn't really matter. But I don't see what in the performances (as opposed to results) is causing the doom and gloom from some posters.

We're doing a lot more right than we're doing wrong. If we continue to do that, the results will come. The concern is that the lack of results starts to have a negative mental impact and that causes the performances to drop off further.

Yup, we've been up considerably in almost every game, and I don't disagree with any of these below, which is why I'm not having a melt down like some on here



Then looking at the table it thinks we should be top, and still has us forecast as 6th based on points accrued only, as in it thinks we've had hard games. So I'm not sure what games others have been watching to think we deserve to be where we are, as it's not been the games I've been watching, or what the stats suggest (which backs it up).
I don't think anyone would disagree that the top 5 there would have been expected to be up there at the end of the season.

 

[BoroMart]

I would largely agree with this. We probably don't have the number of points we deserve at the moment. The flip side is, if this continued, you would start to wonder if it was bad luck or bad judgement. We are nowhere near this point at the moment.

With 3 key and incorrect ref decisions going against us it's clearly at least partly bad luck, those are not in our sphere of influence.

 

[Laughing]

xG is a computer algorithm which has analysed 300,000 shots, from Opta, who are the daddy in stats. It factors in every player's position (with/ without the ball), assist type, header/ good foot/ bad foot, angle, distance, one on one's etc, it's going to be far more accurate than 2 professional analysers or 10 fans watching a game, and maybe similar accuracy to 100 people watching a game, but only because their errors might even out. I don't think it factors in which player is taking the shot, but it shouldn't, as it's not a player measure, it's a measure of chance quality.

It's like on Saturday, I said we were unlucky as that chance they scored probably had a 1 in 50 of going in, it was actually 3%, so 1 in 33, so I was a little off, but I mentioned the guy who scored it hadn't scored for a 100 games, so it wasn't like De Bruyne hitting it. For him to score that was probably a 1 in 100 or more. Then also it should have been given offside as Steffen's view was blocked.

Se below why Mowatt's chance was 10x better, hard to disagree with.

View attachment 43253

It's 5x more reliable than shots/ shots on target, as it's putting a value on each of them which is probably accurate to 5-10%. 10x 25-yarders isn't worth the same as a chance on the penalty spot etc.

It's like me giving you the option of a Penalty (probably 80% chance) or three 25 yard strikes, nobody is going to pick the 3 x 25 yarders, which makes the shots/ shots on target very unreliable unless they all have similar value, which they don't and there's never enough in one game to make it a measure to rely on for that game.

Your example on saturday highlights what is wrong with xG. That is not a 1 in 33 chance. I don't even think its a 1 in 50. It's probably a 1 in 100. It's opinion based.

I know for a fact that shots on/off target are a much better predictive stat than xG. I agree that it doesn't give context on the shot. Did it just miss, was it on target but an easy save, as 2 examples.

I know it is a better predictive stat because I have 10k games analysed with lots of different stats. These are real world before kick off analysis. I know what works and to what degree. Whether I use google, BigML or Azure, or even my own back propogating nueral net, all of their deep net algorithms put a very low weighting on xG when compared to other stats. Time of first goal is more heavily weighted, for reasons I don't know, nor understand.

Opta's stats are fantastic, no argument there. What they don't povide is how influential they are in predictive algorithms.

 

[Laughing]

With 3 key and incorrect ref decisions going against us it's clearly at least partly bad luck, those are not in our sphere of influence.

Absolutely.

 

[Andy_W]

Your example on saturday highlights what is wrong with xG. That is not a 1 in 33 chance. I don't even think its a 1 in 50. It's probably a 1 in 100. It's opinion based.

I know for a fact that shots on/off target are a much better predictive stat than xG. I agree that it doesn't give context on the shot. Did it just miss, was it on target but an easy save, as 2 examples.

I know it is a better predictive stat because I have 10k games analysed with lots of different stats. These are real world before kick off analysis. I know what works and to what degree. Whether I use google, BigML or Azure, or even my own back propogating nueral net, all of their deep net algorithms put a very low weighting on xG when compared to other stats. Time of first goal is more heavily weighted, for reasons I don't know, nor understand.

Opta's stats are fantastic, no argument there. What they don't povide is how influential they are in predictive algorithms.

There's not much difference between what you highlight, 1 in 33 and 1 in 100, it's only 2%, a 2% accuracy would be incredible.

xG is far better as a measure of the quality of chances in a previsous game, so better to reflect how the game should have went if replayed 100 times etc. It's not a good tool to predict the next game, I've never said it was, as nobody knows what chances you're going to get, or which team is going to play, who is injured or who wins the tactical battle. You could create 100 chances against Hull at home, but the next game you might be away at Burnley, you're not getting the same number of shots or chances there, but you might get a higher xG if you get a 1 on 1 or a penalty etc, only that is known after the game.

Better teams will likely have more shots and shots on target, of course they will, but their xG will largely reflect this, after the game.