@Akavall said in #62:
>
I agree there might be many ways to ask the same questino in this big joint probabilities problem, with lots of hidden dimensions and for those dimensions considered lots of hidden values.
I still have to wrap my head around the modelling flow that might be the basis of your confidence assignment definition. I am missing a lot of brackground, but I don't let that prevent me from participating.
Althoug your post seems to be a clear example, and I appreciate the ASCII effort. (integrals and full global problem presentation would be lots of work.. so this "esquisse" is nice to have, to later relate other things I might learn some other time.. Thanks. unless you want to expand or teach me basics about the distribution assumptions, if any, behind being able to compute confidence intervals that way.
But I have that part well cerned (or ecompassed) of my ignorance. so I am good. ( I work from outmost contour going in, in general.. not by fancy, but my nature...). If might not be the place to teach such things. I am fine with that.
It is interesting to use lichess classical rating as that has its pros and cons.
The con is that barely anyone plays lichess classical, as you can see the pool of weekly classical players is only 40K compared to 400K rapid players. I wouldn't have much use comparing my lichess classical to my OTB, but at the same time I already play OTB so I wouldn't need the conversion anyhow.
The pro is that using this probably makes your conversion a lot more accurate than if you considered lichess rapid ratings. I see a lot of people trying to compare their blitz and rapid to OTB using your table, despite you explicitly mentioning that wasn't the data used...did they even read the post?
I imagine lichess classical actually emulates the otb experience in that you can only play a few games in a session, and you probably aren't going to be treating the games super non-seriously. Even still, people who actually enjoy classical (not me) may be facing opponents who berserk them in a classical arena (the berserked player would then get 10+0), so weaker rapid players may get a boost in classical, and people who are strong in rapid wouldn't ever touch classical.
And then the time pressure variant rulesets and their various own layer of variation factor in chess emerging strategies. Different chesses.
And still only pair competition same rating system used that is blind to any chess logic (from the board). So much thinking in words has to be brought to bear on the mathematical models out of such very thin or floating measure system. It appears that having so many categories of chess and possibly soft supported arguments about which part of "trust" and then derive others, would make for a monster full global joint probability problem to eventually get such coveted conversion from various parts.
That those rating systems with dependent hidden variable of their respective pools, might not even be averaging the same skill sets, seems to be a major problem to me. As the previous post triggers in my reading of it. Or reminds me about the underlying competition objects being rated.
@DylanYuIM said in #82:
> Even still, people who actually enjoy classical (not me) may be facing opponents who berserk them in a classical arena (the berserked player would then get 10+0), so weaker rapid players may get a boost in classical, and people who are strong in rapid wouldn't ever touch classical.
The idea of an arena is not very compatible with what people usually want from classical time control. Which is why there are not so many classical arenas and those which exist often disallow berserk. Also, from what I have seen, many users (including me) rather take them as an opportunity to play some games (quick pairing doesn't work well for longer time controls) and do not really aim to score as many points as possible.
@mkubecek said in #84:
> The idea of an arena is not very compatible with what people usually want from classical time control. Which is why there are not so many classical arenas and those which exist often disallow berserk. Also, from what I have seen, many users (including me) rather take them as an opportunity to play some games (quick pairing doesn't work well for longer time controls) and do not really aim to score as many points as possible.
That's a good mindset to have. The main problem with online classical (arenas/pool) is that you don't know how seriously your opponent is taking the game, even if you yourself are treating it as seriously. In OTB you don't have this problem because you can safely assume everyone is trying their hardest because of costs to enter a tournament. I'm glad the Dojo League exists for people who are serious.
I carried out a similar analysis a few years ago comparing ECF grades (as they then wete, i.e. a 3 digit number and a different method of calculation than currently used) and FIDE ratings.
That also seemed to show that the standard formula used to convert between them overestimated what players FIDE rating would be.
But there are a couple of big problems with the comparison that are hard to allow for. Firstly the majority of ECF players don’t play a great many FIDE rated games, so their FIDE ratings don’t have a great deal of opportunity to converge to what their true strength level should be. Secondly because they don’t play a lot of FIDE rated games their current FIDE rating is often actually based largely on results from a a few years ago or more, whereas their current ECF rating is based on much more recent performances. i.e The two ratings relate to different points in time.
The majority of ECF players don’t have a FIDE rating at all, and especially not a classical one. The ones that do are a distinct subset whose characteristics are probably a bit different than the wider player pool.
From what I’ve heard a similar situation holds in the US and in India too where it seems to be widely felt that lots of rapidly improving players FIDE ratings lag far behind their playing strength.
Arguably if the player pools in different countries don’t have much interaction with each other there is not a lot of reason to think FIDE ratings are comparable across countries. An individual’s rating tells you about their relative strength within their own player pool, but there is only a weak mechanism to align different pools with each other.
Personally the only FIDE rated events I’ve played in were a couple of blitz events. I scored around 85-90% across I think 19 games, but that resulted in a FIDE rating not so much higher than the average of my opponents ratings, and nothing like the level that would translate into expecting a 90% score. Of course all my opponents ratings were also fairly unreliable because they also don’t play many FIDE rated games, and may not have done so for years!
I can appreciate the massive effort it took to put this together, but it doesn't take into account how skewed things are because of covid and the stagnation of OTB ratings compared to the massive activity of people online, AND it doesn't disregard provisional ratings, lumping a large portion of the 1300-1500 lichess rating pool into a very low USCF and FIDE rating bracket. It just isn't accurate and contributes to a lot of people feeling unnecessarily bad about their chess journey and how far they've come to see someone just relegated to a beginner status by math which does not discern the nuances like that.
Much love and light to all <3
As a mathematician I was interested to know what language/app you were using for your graphs. I started out writing regression models in BASIC in the 1970s! Thanks for your info.
@NoseKnowsAll said in #74:
> @MBurns2020 Because you have all of your profiles linked together, I was able to explore you specifically. Your lichess classical = 2030, chesscom rapid = 2100, and FIDE = 1900. Based on my converter, I would guess that both your chesscom rapid and FIDE seem accurate as they are quite close together, and you are underrated on your lichess classical rating. Indeed, when I look at your FIDE plot, you've gained a lot of rating this past year (200+ FIDE in one month! Nice job), and yet you haven't played a lichess classical game in months.
>
> So I would disagree with your assessment actually. I would say your are almost perfectly modeled by my converter. Play some lichess classical games and I would guess you could make it to 2200 in no time!
hmm, ok thanks for the correction