OpenAI's team defeated several top players in it, but lost even against a retired team of professionals in 5 vs 5 despite starting the games stronger (but weren't able to combine the strategies as well as humans). DeepMind's defeated a couple of Grandmasters in Starcraft.
I will give my batchelors and masters back then. xG isnt strong enough apparently
Hmmph okay not really disagreeing with your initial point but they were heavily beaten by the retired team as far as I know.
Regression will happen anyway, we simply won't reach 102 points next season, but the effect would be much smaller than now.
That's what I said. They lost when played 5 vs 5. Each bot was playing better than a human, but not when joining forces. We are still more creative at some tasks. Still.
I think we're getting confused here. There was an "alternative league table" that corrected what the scousers saw as injustices. Every time a liverpool player went down in the box without getting a penalty, they added a goal on to the scoreline. Everytime we scored a goal, if we got a questionable throw-in 5 mins earlier, they took that goal off. Then they put these adjusted scores into the league table and declared themselves moral champions. (only a very slight exaggeration on my part there)
They are models done by different companies. It is machine learning, not some analytic solutions. Some models will be great, some crap, those that crap will disappear, the good ones will be sold for millions to teams.
Just wanted to make the point that overperforming isn't that big of a deal over a limited amount of time. I was just giving Pogue an answer that it actually isn't that grim. In fact Tottenham are overperforming more so than we are since Ole has taken over.
I have been iin RAWK for years, and the only alternative table there has been a similar version (if not identical) to Storey's table. The alternative table which shows that they win the league didn't ever exist in reality (at least not in RAWK). It is just a Mandela effect.
Hm, I'm sure I've seen it hosted elsewhere (not actually on RAWK) but until I can find it I'll have to hold off.
Most of these numbers make sense, but something stand out. Barcelona only 7.7% up with Messi being so good is low. I guess Suarez have missed far too much to reduce those numbers.
Nonetheless it can create confusion if you label it all xG. That 11tegen11/between the posts bloke on twitter for example had a figure for our Crystal Palace game, which seemed absurd.
Not arguing anything you mentioned. Just adding a tangential point that in terms of factors, some models such as Stratabet's model factors in positioning of defenders as well. So results will vary from models to model. I am not sure the model posted by OP factors in defenders. I reckon the models that factor defenders' positioning are better indicators than the ones do not.
What confusion it is? It has always been like that that some models are good, some are bad, despite trying to do the same thing. It isn't even special to data analytics, Google gained ground cause their search engine was so much better than the others. But all were 'search engines'.
Most of that stems from the title-winning season 16/17. The other four season I included in the data sample produce numbers very close to xG.
Not mine. I've been defending this stuff for a while.
Defending it why exactly? What good does it do?
Yeah I remember them being very efficient under Conte. Very high confidence and just going so strong over christmas. Although would have expected that too when Mourinho won it. When they did struggle in 15/16 I would have expected them to underperform a lot with Hazard and Costa being in poor form.
Different point of view. Stats show tendencies for a bigger picture usually, which I am really just interested in.
To pick some of the biggest outliers:
What is clueless is this post. Yes, football is much harder to analyse that way than sports that largely consist of set pieces. But everything can be translated into a number (even Klopp's wind), just because we don't know it yet and the science isn't advanced enough at this point doesn't mean it can't be done. This is really such an ignorant view to hold in the 21st century.
De Gea and Oblak are two best keepers in the world. Also, Mourinho (not the shit version we get in third year) and Simeone are great at defense.
I totally agree and it explains why Atletico for example consistently outperform their defensive numbers. But even Oblak and de Gea aren't that much better than the average. That was a freak season even by their standards if we look at their long-term averages.
Oh, absolutely. I don't watch Oblak every week but last season was a freak season for De Gea.
Yup, the widespread view that luck and referee mistakes even themselves out over the course of a season is more wrong than right.
I would honestly rate Oblak as the best in the world. It's outrageous at times what he saves. Atletico really have been blessed with good goalkeepers. Or good scouting I guess.
Well, in the long run they do (unless there is a real bias for some reason). That's just not much comfort when you lose the title on goal difference for example.
Absolutely. The luck and referee decisions balance themselves over the course of a decade (proxy for a very long time), but 38 is not a very big number to say so.
But it's human beings thats involved here. Last year you told me we wouldn't finish second because of xG and around January you said we were nowhere near Spurs and Chelsea for top 4 due to your stats.
I don't dispute that. But all of the factors you listed have a number and probability attached. Some of them we just don't know or have no way of accurately measuring them, others we can approximate. When Harry Kane is replaced by Llorente in Spurs' next game we have a rough idea of how that would influence their chances of winning. But Llorente might play a blinder that day (which again has a probability that we can't pinpoint exactly). xG at least is a very useful tool to filter out some of the noise. Nothing more, nothing less. And United's defensive numbers from last year are up there with some of the biggest statistical outliers. This year they have conceded roughly as many as they 'should'.
Of course they are not, it is an idealized model in the end, not real world. It is far better than the traditional stats (number of shots, possession etc) and in the long term a very good model should outperform experts. I would be totally surprised if within 10-15 years a large part of scouts won't be data scientists.
