Well, niking, I am a little above 1500 at
chess.com, and a little above 1700 here. Although I didn't know that
chess.com changed their provisional rating from 1200 to 1500, I haven't played there since I found lichess.
While ratings are relative, and not at all absolute, I think it's ridiculous to say that they are totally meaningless. ie "THOSE NUMBERS DON'T MEAN ANYTHING!"
I didn't say that those numbers don't mean anything. I said that average ratings don't mean anything (for purposes of comparison across sites).
Well in the way we are dicussing it those number don't mean anything. Unless we have some sort of cross-platform statisctics, we can't truely get anything out of that data hence THOSE NUMBER DON'T MEAN ANYTHING! :D
Gotcha.
the numbers show us that the glicko-system works like it should ;-)
No, I am pretty sure there is nothing which 'works' with this system. Each time, I beat a player with 3-600 points more than me, I should be getting definitely 70-80 points. Regardless if I have played X matches or 5X matches. Anyway, I have noticed the fluctuation (differential) converges towards +- 10 as you play more games.. but still... I play 2 matches with ~1500 say, beat someone 1950, I get +24ish, then lose two tight games to someone 1485 and 1515, and I lose more there than beating someone 3 to 400 "ratings".. In ELO this would be absurd and in reality it would be insulting.
So unfortunately, the system does _not_ work. I beat several times people above me by hundreds, lose to someon 100 less too a match. I am one of those famous no consistency whatsoever, which allows me both to be a great player, an average player and a piss poor one; not only one of them.
and the system fails epically when dealing with cases. So either improve the maths or don't use dysfunctional rating systems please. It has made the site extremely unprofessional (although with all the variations it has become more popular for sure).
Popularity is nice, professionality in chess is always expected to a degree though, so please.. use ELO or something which actually works. (I have many log ins but is irrelevant, but for those thinking this profile was relevant).
#26, the reason it doesn't do that is because Glicko-2 is actually a much better system than Elo. That's why virtually every forward-thinking chess federation and website have adopted it. Which is why I find it particularly bizarre that you claim the site is 'unprofessional' for it.
The problem with what you said is that you don't take account of the Rating Deviation, e.g. ±300, which I strongly feel should be displayed by lichess. The rating system is built so it gets a more accurate picture of your rating the more games you play.
The reason you don't get a lot of points when you beat players much higher-rated than you is because you can't beat them *reliably*. When you can win against them regularly, the rating changes to reflect that. But in Elo, you lose on a bad day and it's a grinding journey to go up again.
It's not about satisfaction, it's about fairness. And that's what Glicko-2 has that Elo doesn't. It is worth noting that Glicko and its successor system, Glicko-2, have been introduced relatively recently compared to Elo, and yet have very wide adoption. There are people who are much smarter than us when it comes to mathematics and designing such systems in general. I feel they made the right decision by dropping Elo in favour of this.
lolcat, please read Prof. Glickman's research paper at
http://www.glicko.net/research/glicko.pdf and advise -- beyond the professor's own criticism about the time-sensitive nature of the system -- why you think Glicko doesn't work.
I'm trying to figure out how Glicko-2 improves over Glicko, but given its widespread adoption on most ICSs and many gaming servers in general, one must assume that Glicko-2 is an improvement and not simply an attempt to extract licensing fees (since Glicko is public domain but maybe Glicko-2 isn't?).
#28, they are both in the public domain as Glickman's website indicates:
http://glicko.net/glicko.htmlThey are both made out of goodwill to produce the best rating system possible.
Toadofsky.. I would like to rather be reminded of all the obviously poor things ELO had then? I seem to not recall, and felt it was perfectly adequate and had a logical, consistent spread to it. it has bene utilised a long time in the chess ocmmunity but perhaps it has glaring mistakes I just wasn't aware of.
Again I beat someone with 450 rating more or so, they lose 27, something I lose seemingly at random to someone around same ratings as me lol.. In 3 matches it's won back by him I saw, still beating some whih are 200 less, and ye.. it's a useless system, a paper on numerical analysis or stochastics and game theory is fine and all, but the very point was, it doesn't correlate well to real gaming of chess. So, I do not doubt the paper is mathematically valid, presuming it has been peer reviewed since one has implemented here, but it just ain't good. :) Hence, bring back elo please so we have something proper to relate to.
