Whole History Rating
|Table of contents
- Whole-History Rating: A Bayesian Rating System for Players of Time-Varying Strength, Conference on Computers and Games, Beijing, China, 2008.
- Mailing list announcement and discussion of Whole-History Rating
WHR is an improvement on the decayed history algorithm (the type KGS uses). Both constantly recalculate the most likely rating for players based on past results. The major difference is WHR takes into account player's strength changes with time, and calculates a separate rating for each day someone played at least one game. Decayed history attempts to deal with player's strength changing by lowering the weight of older games, but this does not work as well.
yoyoma I made an implementation of this algorithm here: source (see the whr directory). Since I'm interested in comparing this to the KGS algorithm, I included code to do transforms from vanilla Elo numbers to kgs numbers. For the examples below the transforms are for 2d+ players (1 rank = 226 Elo, KGS uses slightly different constants for kyu players).
Assume you played X games a day as an average dan player of your rank for 180 days. Then you get inspired and play a whole bunch of even games (still with average players of your rank) in one day and win them all. How many wins does it take to promote 1 stone?
|After a year, one game changes your rank by ~1.3 stones
GPD - Games Player per Day, for the 180 days before the break+winstreak Break - Number of days of no playing after the 180 days of constant play Games - Number of wins in a row needed to promote one rank (half a stone)