Index

Welcome to KGS
Downloading and Installing CGoban, the KGS Client
CGoban Main Window
Introduction to KGS
KGS Plus
Tournaments
KGS on Androidâ„¢
Terms of Service
Admin Announcements
Newbie FAQ
Main KGS Window
Set Preferences
Server Stats
"New Game" Window
Game Window
Playing Games
Game Types
Supported Rule Sets
Supported Time Systems
Tagged Games
Rating System
Rating System Math
Main FAQ

Information for Help Editors
Resources for Help Editors
Help Discussion
Recent Changes
Translations Available (click a flag to view):
English (United States)

Help Page Archive

(code snippets)

Short sections of help pages can be archived here. To archive an entire help page, put a link to the file containing the page in the Archive Index.

Archived from "How to Get a Game". dj 2/24/2011

Automatch can help you get rid of a question-mark ranking because automatch ignores question marks when making a paring [probably not true]. Be sure not to click "Free" in the automatch preferences. If automatch takes too long to find you a match, you can speed it up by adjusting the parameters.

The Independence Assumption in the Rating System Math. dj 9/20/2010

I couldn't find a good reference for this analysis. Perhaps my memory regarding the key point fails me. I believe that this analysis is valid, but, I am not willing to put in in the help pages without confirmation.

4 It is not clear to the current author (dj 9/10) whether the independence assumption is a problem or not.

One possibility is that if a player plays significantly more games against stronger players than they do against weaker players, then their true playing strength is stronger than is indicated by their rank. In other words, it is possible, perhaps even likely, that the independence assumption builds an artificial "sandbagging" factor into the ratings of tilde (~) players.

Ironically, if the algorithm builds a "sandbagging" factor into tilde (~) players, then it also builds in a sandbagging factor for players who play significantly more of their games against weaker players. Players are exempt from this effect if they play most of their games players of the same strength. For the technical details, see the next footnote.

5 This method of computing probA assumes that a player's win/loss results are independent.

In practice, this method often yields values for probA that are inappropriately small. [KEY POINT] This results from the multiplication of small values of PA wins or of PA loses . If a player plays lots of their games against stronger players, then the algorithm artificially inflates their rank to compensate for this effect. Players with lots of games against weaker players also have a "sandbagging" factor built into their rank.

This effect is stronger when most of the games have small rank differences. If most of the games were played with about the same rank difference, the effect would disappear, even if the rank differences were large.

There are commonly used ad hoc techniques that keep probA from getting too small. The current author (dj 9/10) does not know how effective these techniques would be if they were used in KGS. It is likely that their effectiveness in general depends on the specific environment in which they are applied. It is probably not a good idea to try them in KGS because of the time-honored principle, "if it ain't broke, don't fix it."

Related Links


Edit this page (requires admin or translator privilege)


KGS