Rating System Broken.

@SavageAntarctican

It is the shame there is such people claiming the implemented rating system is perfect here. It is quite opposite. You probably didn't understand anything wrote in this thread, nor understant the Glicko-2 rating system.

http://www.glicko.net/glicko/glicko2.pdf

Try to read and understant it fully, I suppose you have some basic math background. Then get the parameters Toadofsky point in source and analyze it, then comment and make accusations.

I can claim that no body undestand Glicko-2 fully, just use it blindly! Or deliberately make parameters in such manners that all wrote here about it ( that is just useless), is simply plain fact.

It is the shame that many of you here are so arrogant that you simply neglected plain facts and start apriori with personal accusations and insults, instead to focus to fix the real issue here. That is what is unbelievable! Then, simply there is no point to argument with you about anything further.

funny pic
CANT LEAVE LOWER 0RBIT
🤣

@n321 First of all, I explicitly told that no rating system is perfect. Then, I showed you a few examples on why Glicko-2 is best suited for lichess. Finally, I told that if you are able to find a better rating system, then please tell me. There aren't that many rating systems in the first place.

What else do you want?

There is no issue here, just pointless agrument that goes round and round in circles. And you just try to continue the argument and say that you don't want to argue? Next time, please don't be hypocritical and nonsensical at the same time.

@Toadofsky itself pointed several issues here (his post here about tau parameter etc), what else to point?

I have no time nor will to examine Lichess implementation of the algorithm from sources in order to evaluate it.

Involving probability in rating system is hardly good idea, as I have mentioned here many times. At least minimizing their effect can be considered as a probable solution for already implemeneted system. Mixing statistic and probability here is mixing apples and oranges, oil and water - that is absolutely pointless, as again I have mentioned here already several times.

I'm also not willing to examine other existed rating systems. It is similar, if involves probability, similar problems will happens.

We only agree that is pointless further argumentation as any of this will probably never be taken in consideration to be changed. Waste of time.

@Toadofsky

From another thread: "If you want to have an argument, please... sorry, I don't know how to politely put it."

Oh, just close this thread as well if you do not like all this, you can do that, no doubt. That is the way to go, this is your (Lichess) "little kingdom" and you may do what is please to you, rest must blindly obey.

Feel free to close my account as well.

Some guys are investing a lot of time writing about how the rating system is broken because it awards them only so few points for every win. Judging from the fact that in the same time they are bitching and whining here but they didn't get better I would say they belong exactly at the rating level where they are.

If you deserved to be 100 or 200 points higher you already would have gotten there in the meantime. The fact that you have not... well, go figure.

#155 I didn't close "One of the worst pairing systems!" because I believe the Streisand Effect is a thing.

I've been reading this topic although ironically the strongest argument I've seen is what you posted in the other topic, sharing some evidence. My two conclusions there are logical:

1. You are mostly playing lower-rated players (playing in tournaments may change this). lichess.org/stat/rating/distribution/blitz shows that once players reach a century (an even hundred number of points) they stop playing, which may explain why players 1500-1525 choose not to participate in the lobby? You don't want to hear this, but if you are 200+ points stronger than a 1400, you should win 76% of the time... but my main point here is that players rated over 1500 must be absent in the pairing pool if you keep getting paired that way, so tournaments might be a better route to finding competition. We can't force over-1500 players to play in the lobby if they're choosing to withdraw from competition; that's not a flaw of the pairing system.

2. Accurate rating models have some sort of mathematical basis, e.g. TrueSkill, Elo, Glikco, or even the US Chess Rating System (or others). It doesn't make sense to argue that a player has 1900 strength and then discredit the math as to what 1900 means.

I'm actually a bit mad at whoever closed that topic since you made a good point:
"I'm still playing currently what computer choose with quick pairing."

Combined with what we already know from lichess.org/stat/rating/distribution/blitz , assuming other rating proposals all fail there may be value in Lichess developers carefully studying the quick pairing algorithm with much more data than I am considering right now, and see if there is some way to improve that pairing algorithm. Again, we can't force >1500 players to play if they choose to withdraw altogether, but *if* there is a flaw with the pairing algorithm (and not just a problem that those players are choosing to play tournaments instead of the lobby) maybe it's possible to improve that. This doesn't negate that a "1600 strength" player should have a 76% win ratio over 1400-rated opposition, etc., but *if* somehow >1500 players are choosing to play but can't find similarly-rated opponents, with tremendous effort it may be possible to find a way to improve the pairing algorithm.

Then again, I am hopeful that my 2 proposals (both of which require diligent research; work and life and github.com/ddugovic/Stockfish/issues/574 and dealing with other forum topics are exhausting my effort) will help:
* There is a nonzero advantage to moving first in chess
* Glickman suggests that hyperparameter tau should be tuned based upon predictability of game outcomes, yet we are assuming that ultrabullet, bullet, blitz, rapid, and classical outcomes have the same outcome distributions (and maybe they do!)
Comic: "Subjectivity" xkcd.com/255/ (it's not yet on imgur and I lack permission to publish it there)

#155 seenms to be playing mostly unrated games and hence opponents vary. But is someonest rating is 1500 he is bound to get players around his rating as thats where most players are. In blitz I am aroudn 1500 get opponent in matter of 10 seconds at most

Yeah, but also when I filter for his rated games, his explanation makes some sense. There's not enough evidence yet to support completely overhauling the pairing system, but evidence suggests that further study may be merited.

I am not sure @n321 prefer the ranking system. e.g, You can make 30 rankings by equally splitting 30 million players into 1 million each in each ranking category. However, in current Glicko 2, the rating system is Pyramid shape, with massive tapering in two ends!