What is the logic of lichess rating giving system ?

Back to earth and the op title (not the post, necessarily).

So the game start ratings are not 100% determining of the increment/decrement distributed to the players post-game outcome?

I have been confused in my case in correspondence, with many games in parallel, staggered starts, overlapping games of different lengths, as to which rating instantaneous might have been used.

But otherwise, would volatility (SD or RD whatever it is being called here) have an effect on that. Perhaps from glicko1 to 2 this happens? I may have forgotten that, but I thought the rating interactions with other player kept avg and deviation separate.

I would appreciate a refressher (being lazy about reading out of discussion, or having little attention or energy, etc...)

Back to earth and the op title (not the post, necessarily). So the game start ratings are not 100% determining of the increment/decrement distributed to the players post-game outcome? I have been confused in my case in correspondence, with many games in parallel, staggered starts, overlapping games of different lengths, as to which rating instantaneous might have been used. But otherwise, would volatility (SD or RD whatever it is being called here) have an effect on that. Perhaps from glicko1 to 2 this happens? I may have forgotten that, but I thought the rating interactions with other player kept avg and deviation separate. I would appreciate a refressher (being lazy about reading out of discussion, or having little attention or energy, etc...)

dboing

edited

#52

https://lichess.org/page/rating-systems
https://lichess.org/faq#ratings

Basically, the system is 95% sure that your rating is somewhere between 500 and 2500. It is incredibly uncertain. Because of this, when a player is just starting out, their rating will change very dramatically, potentially several hundred points at a time. But after some games against established players the confidence interval will narrow, and the amount of points gained/lost after each game will decrease.

https://en.wikipedia.org/wiki/Glicko_rating_system#Step_1:_Determine_ratings_deviation

After a game, the amount the rating changes depends on the RD: the change is smaller when the player's RD is low (since their rating is already considered accurate), and also when their opponent's RD is high (since the opponent's true rating is not well known, so little information is being gained). The RD itself decreases after playing a game, but it will increase slowly over time of inactivity.

The wikipedia article is clear about the influence of the RD=SD (rating or standard deviation) on the calculartion of increment/decrement per game event (post-game) given the pair of estimated ratings and their deviations prior to game.

so yes it is possible if your standard deviation is large or not that the same given rating pair will not give the same rating gain or loss for same outcome. for the amount of influence one would have to look at source code or get an idea form the wikipedia math (or the original statistical papers, or the computation blurb versions).

In any case I think Lichess may have to make some parameter choices about characteirtic game frequencies per time control "variants" (Does it?). Or does correspondance get the same volatiility increase per unit time as bullet?).

But the main idea is that the rating estimate uncertainty will affect the gain/loss per game.

looking at the equations it seems that there is no adjustable parameter (wikipedia equations).
in any case, for getting to the source
http://www.glicko.net/glicko.html

Some people read better a computed example, I prefer the more abstract math. I find the Wikipedia and the statistical papers more informative than the often referred to example of calculation, with lots of numbers cranking. To each its own.

https://lichess.org/page/rating-systems https://lichess.org/faq#ratings >Basically, the system is 95% sure that your rating is somewhere between 500 and 2500. It is incredibly uncertain. Because of this, when a player is just starting out, their rating will change very dramatically, potentially several hundred points at a time. But after some games against established players the confidence interval will narrow, and the amount of points gained/lost after each game will decrease. https://en.wikipedia.org/wiki/Glicko_rating_system#Step_1:_Determine_ratings_deviation >After a game, the amount the rating changes depends on the RD: the change is smaller when the player's RD is low (since their rating is already considered accurate), and also when their opponent's RD is high (since the opponent's true rating is not well known, so little information is being gained). The RD itself decreases after playing a game, but it will increase slowly over time of inactivity. The wikipedia article is clear about the influence of the RD=SD (rating or standard deviation) on the calculartion of increment/decrement per game event (post-game) given the pair of estimated ratings and their deviations prior to game. so yes it is possible if your standard deviation is large or not that the same given rating pair will not give the same rating gain or loss for same outcome. for the amount of influence one would have to look at source code or get an idea form the wikipedia math (or the original statistical papers, or the computation blurb versions). In any case I think Lichess may have to make some parameter choices about characteirtic game frequencies per time control "variants" (Does it?). Or does correspondance get the same volatiility increase per unit time as bullet?). But the main idea is that the rating estimate uncertainty will affect the gain/loss per game. looking at the equations it seems that there is no adjustable parameter (wikipedia equations). in any case, for getting to the source http://www.glicko.net/glicko.html Some people read better a computed example, I prefer the more abstract math. I find the Wikipedia and the statistical papers more informative than the often referred to example of calculation, with lots of numbers cranking. To each its own.

dboing

edited

#53

for a series or sequence of m games, where (there is no markdown here to replace the latex)

r_{i} represents the ratings of the individual opponents.
{\displaystyle RD_{i}} represents the rating deviations of the individual opponents.
s_{i} represents the outcome of the individual games. A win is 1, a draw is {\frac {1}{2}}, and a loss is 0.

E operator stands for expectation. and the dependent variables are the various games opponents parameters, and the game outcomes. This can be done for one game too. or any consecutive sequence.

Best is look at first link for this post.

The thing I don't understand is that it appears to assume a clock of updating. The m games do not seem to have had new rating estimated within. perhaps m is the clock. can anybody look and confront?

Or it is as if the m games (m opponents) were done simultaneously. that r0 was the starting rating for each games. Perhaps finally the example might be better (harder to read though).

https://en.wikipedia.org/wiki/Glicko_rating_system#Step_2:_Determine_new_rating I would have to put the following image on imgur, and then refer to that.. forgot how... to embed. https://wikimedia.org/api/rest_v1/media/math/render/png/ab5c89b028e21369fbe39c00671715e4c1626d59 for a series or sequence of m games, where (there is no markdown here to replace the latex) >r_{i} represents the ratings of the individual opponents. >{\displaystyle RD_{i}} represents the rating deviations of the individual opponents. >s_{i} represents the outcome of the individual games. A win is 1, a draw is {\frac {1}{2}}, and a loss is 0. E operator stands for expectation. and the dependent variables are the various games opponents parameters, and the game outcomes. This can be done for one game too. or any consecutive sequence. Best is look at first link for this post. The thing I don't understand is that it appears to assume a clock of updating. The m games do not seem to have had new rating estimated within. perhaps m is the clock. can anybody look and confront? Or it is as if the m games (m opponents) were done simultaneously. that r0 was the starting rating for each games. Perhaps finally the example might be better (harder to read though).

dboing

#54

So how would one go about manipulating that, artificially?

thinking with #52, uncertainty gives bigger swings. and uncertainty increases with time spent not playing (or between games).

so playing not often, would increase a win gain, even over a lower rated opponent. But the furthest the lower opponent still the smaller (examine the formula, the distance is in the denominator, pretty sure I saw that).

The hopping manipulator would have to keep winning within same difference lower rating to keep having same gain.
And the probabilities are that the manipulator would still have some losses, and then bang in the face of the manipulation.

So how would one go about manipulating that, artificially? thinking with #52, uncertainty gives bigger swings. and uncertainty increases with time spent not playing (or between games). so playing not often, would increase a win gain, even over a lower rated opponent. But the furthest the lower opponent still the smaller (examine the formula, the distance is in the denominator, pretty sure I saw that). The hopping manipulator would have to keep winning within same difference lower rating to keep having same gain. And the probabilities are that the manipulator would still have some losses, and then bang in the face of the manipulation.

dboing

edited

#55

That was for the artificial boosting manipulation.

The sandbagging manipulation can't be avoided under any rating system, and if you see a good move, doing another move at random is likely to make you lose some game. It is easier to lose I guess.

So, perhaps the claims of manipulation might be about combining some subtle sandbagging on higher/lower opponents in order to aim at a hundreth rating within the manipulator true chess skill reach with frequent enough game to have some control in the small changes in the needed directions.

I say hundreth, but it could be any choice within abilities of player. I don't know how refined such sub-clinical sand-bagging could occur (it might not be sand bagging in the sense of trying to get in some tournament, but just for the sense of power over own rating...).

People getting bored with chess, I guess. (so here not about inflated rating).

That was for the artificial boosting manipulation. The sandbagging manipulation can't be avoided under any rating system, and if you see a good move, doing another move at random is likely to make you lose some game. It is easier to lose I guess. So, perhaps the claims of manipulation might be about combining some subtle sandbagging on higher/lower opponents in order to aim at a hundreth rating within the manipulator true chess skill reach with frequent enough game to have some control in the small changes in the needed directions. I say hundreth, but it could be any choice within abilities of player. I don't know how refined such sub-clinical sand-bagging could occur (it might not be sand bagging in the sense of trying to get in some tournament, but just for the sense of power over own rating...). People getting bored with chess, I guess. (so here not about inflated rating).

dboing

edited

#56

So the reasoning stays the same as based on that.

I am sorry, with my internal flipper debating thingy, i tend to flip things around very easily...

Enfer et Damnation: I did it again. trying to be combining both increment and decrement might be the problem, trying to save some typing. The higher rating losing loses more than the lower rating losing, but that increment or decrement (as applicable to both lower or higher and winner or loser, do the combos), is in absolute value an increasing function of the rating distance or difference (which ever way). (and according to text also increases with hiatus duration).. hope I did not flip anything.. so many words.... latex (old mathML) would be less flippant.

I may have inverted the polarity of a function monotonocity (increasing versus decreasing). the equations don't matter. They should show what the text is saying. The rating increment/decrement is an increasing (in absolute value) function of rating opponent distance to own rating, and an increasing function of time between games (or sets of m games, if there is an update game # clock). So the reasoning stays the same as based on that. I am sorry, with my internal flipper debating thingy, i tend to flip things around very easily... Enfer et Damnation: I did it again. trying to be combining both increment and decrement might be the problem, trying to save some typing. The higher rating losing loses more than the lower rating losing, but that increment or decrement (as applicable to both lower or higher and winner or loser, do the combos), is in absolute value an increasing function of the rating distance or difference (which ever way). (and according to text also increases with hiatus duration).. hope I did not flip anything.. so many words.... latex (old mathML) would be less flippant.

This topic has been archived and can no longer be replied to.