Not the one with Simon Moore between the sticks tho!Just kick the ball into the goal...
It looks pretty accurate in relation to general positions in the tablexG is bollocks as a stat.
xG is bollocks as a stat.
Do we think our high xG might be slightly inflated based on the opposition we've faced so far?
Would be interested to hear of stories where a team that had a really great xG at the start of the season without scoring too many, then turned that into loads of actual goals later on, mainly just to make me feel better about Saturday...
Yes your xG would still be less than 1, as it's based on expected outcomes out shots rather than what actually happened.I have a question for people who follow this...
If you are playing Man City & getting battered with no attacks or shots on target... But only for a CB bringing the ball out of defence & hitting a 35 yard worldie into the top corner. Would the xG be less than 1 (what they have scored)?
Why the fuck did we sign Waghorn?
I will never be in a pub talking about our xG, but will more likely cursing the chances we had & missed... I guess this kinda formalises this, which is fine.
I have a question for people who follow this...
If you are playing Man City & getting battered with no attacks or shots on target... But only for a CB bringing the ball out of defence & hitting a 35 yard worldie into the top corner. Would the xG be less than 1 (what they have scored)?
I suppose if a striker who scores 20+ goals in a season in a team with a low xG may be seen in a better light.
Do we think our high xG might be slightly inflated based on the opposition we've faced so far?
Would be interested to hear of stories where a team that had a really great xG at the start of the season without scoring too many, then turned that into loads of actual goals later on, mainly just to make me feel better about Saturday...
Why?
xG is bollocks as a stat.
Why?
Imagine a midfielder playing with awful strikers who keeps putting the ball on a plate only to watch the donkeys fail to hit a cows arse with a banjo. xA would be high even though assists are low, which might tell you he’s worth a punt in your team with decent strikers. Same the other way around, a striker with low goals but also low xG just isn’t having chances made for him.
I'll have a shot here.
You are relying on in-game statistics instead of in-game heuristics
It is a bad way to try to capture the uniqueness of position.
We had an xG of around 1.5 against the Loftus Road Galacticos
Who is relying on anything?
It’s just another data point. More data is always better.
Machine learning has proved the value of statistics with big enough data over heuristics.
That's not correct.
xG is like putting winning percentage together to find your best move instead of using engine evaluation to measure heuristics.
It really is bad sports science.
Evidence is below.
Chess Opening Explorer - 365Chess.com
With our Chess Opening Explorer you can browse our entire chess database move by move. The Opening Explorer is the best tool if you want to study chess openings.www.365chess.com
Chess is a really poor analogy though, it’s got very limited set of states. Football is far messier and harder to model. Modern data science methods like ML are far more appropriate than heuristics which rely on an oversimplified model. This is why your player model based on team performance doesn’t work. It’s too low resolution.
Feature selection and refinement based on big data inputs is far more likely to lead to insights.
Evidence is below
AlphaGo: The story so far
You’re trying to run before you can walk, talking about “moves” rather than individual player actions.
The chess idea could perhaps work if you were able to define a footballer’s ‘move accuracy’ as you can for a chess player.
Shock!Schmmeee, you are wrong.
Schmmeee, you are wrong.
The difference between a standard engine and Alpha-go is how they approach measuring heuristics. One being top-down and the other bottom-up.
This is how they measure a position. They are not relying on previous positions if you take out the opening book and end game database.
xG relies on looking at previous positions which are built on game heuristics instead of looking and evaluating the heuristics in themselves which have built the positions.
You can see how a standard engine evaluates heuristics below.
Thinking Machine 6
www.bewitched.com
You’re missing my point. AlphaGo couldn’t rely on standard heuristic methods because the game space was too large to brute force like chess. Football has an even bigger game space. More akin to something like StarCraft (another game DeepMind are having more success than most at modelling).
Rules based engines like we use for chess will only get you so far, then you need statistical models.