Ranking systems: Difference between revisions
No edit summary |
(elaborate on the earlier ranking systems) |
||
| Line 1: | Line 1: | ||
There have been two basic systems used recently in FreeAllegiance to attempt to rank players with a third on the way. | There have been two basic systems used recently in FreeAllegiance to attempt to rank players with a third on the way. | ||
[ | ==ELO== | ||
This was the first statistical rating system that was tried in Allegiance. I think it was originally conceived and implemented by Your_Persona. The basic ELO rating system is [http://en.wikipedia.org/wiki/Elo_rating_system well described on Wikipedia]. However, ELO was developed for games of 1 vs 1, like chess. To use it for Allegiance, a team's elo points was simply the sum of all the players' elo points on that team. Then post-game adjustments were calculated using the ELO equation and applied to all players. | |||
One major downside of ELO is that it's a zero-sum game. What one player gains, another player loses. Since we have a constant influx of newbies who might lose a few games and then leave forever, this meant that ranks would gradually drift over time. Also, Alleg's ELO system became more and more complex when dissatisfied players tried to "fix" it. Newbies actually started with rank 15 (1500 ELO points), but the displayed rank in-game was reduced by a number based on how long they played. Players ranked as expert and above had penalties that kept their ranks artificially low. | |||
[[AllegSkill]] | ==HELO== | ||
In a bold move to simplify the rating system issue, Pook implemented his own version, based on the ranking system of HALO (1? 2?). It's detailed in [http://www.freeallegiance.org/forums/index.php?automodule=blog&blogid=4&showentry=198 one of his blog posts]. You can read more about the HALO EXP ranking system [http://www.uglyfellas.com/h2ranklogic.htm here]. | |||
This is not a zero-sum system, which means that ranks are more stable over time. However, it has other flaws. When calculating the adjustment of Bob the pilot's rank after a game, it only looks at the players on the opposing teams. It doesn't take into account what other pilots were on Bob's team. This makes the HELO ranks less accurate. | |||
In addition, HELO ranks are not guaranteed to be mathematically sound. How can we know that one inter. 6 is worth twice as much as a novice 7 when autobalancing? This issue could be mitigated by studying the outcomes of autobalanced games and carefully adjusting the rank thresholds (the right-most conversion table on the [http://asgs.alleg.net/asgsnet/leaderboard.aspx leaderboard]. | |||
==[[AllegSkill]]== | |||
This is a proposed replacement for HELO that is developed by Sgt_Baker et al.. It is based on [http://research.microsoft.com/mlp/apg/trueskill.aspx Trueskill], which was developed by Microsoft Research. | |||
==See also== | |||
[[Why is HELO broken and can it be repaired?]] | |||
Revision as of 18:46, 12 December 2007
There have been two basic systems used recently in FreeAllegiance to attempt to rank players with a third on the way.
ELO
This was the first statistical rating system that was tried in Allegiance. I think it was originally conceived and implemented by Your_Persona. The basic ELO rating system is well described on Wikipedia. However, ELO was developed for games of 1 vs 1, like chess. To use it for Allegiance, a team's elo points was simply the sum of all the players' elo points on that team. Then post-game adjustments were calculated using the ELO equation and applied to all players.
One major downside of ELO is that it's a zero-sum game. What one player gains, another player loses. Since we have a constant influx of newbies who might lose a few games and then leave forever, this meant that ranks would gradually drift over time. Also, Alleg's ELO system became more and more complex when dissatisfied players tried to "fix" it. Newbies actually started with rank 15 (1500 ELO points), but the displayed rank in-game was reduced by a number based on how long they played. Players ranked as expert and above had penalties that kept their ranks artificially low.
HELO
In a bold move to simplify the rating system issue, Pook implemented his own version, based on the ranking system of HALO (1? 2?). It's detailed in one of his blog posts. You can read more about the HALO EXP ranking system here.
This is not a zero-sum system, which means that ranks are more stable over time. However, it has other flaws. When calculating the adjustment of Bob the pilot's rank after a game, it only looks at the players on the opposing teams. It doesn't take into account what other pilots were on Bob's team. This makes the HELO ranks less accurate.
In addition, HELO ranks are not guaranteed to be mathematically sound. How can we know that one inter. 6 is worth twice as much as a novice 7 when autobalancing? This issue could be mitigated by studying the outcomes of autobalanced games and carefully adjusting the rank thresholds (the right-most conversion table on the leaderboard.
AllegSkill
This is a proposed replacement for HELO that is developed by Sgt_Baker et al.. It is based on Trueskill, which was developed by Microsoft Research.
