Page 2 of 3
Posted: Sat Sep 02, 2006 6:44 am
by Terralthra
tmc wrote:QUOTE (tmc @ Sep 2 2006, 02:01 PM) But an autobalance button would not prevent this problem greator, since the autobalance would also see everyone as equal.
What you're referring to is a problem deeply rooted within ELO, and has nothing to do with this particular reset.
They would only be able to stack for a couple games each. I went through it in a thread in gameplay, but basically it would not take more than 10-12 games for each player (running in parallel, naturally) before the gains for the more skilled players and the losses for the less skilled players prevent them from stacking and allow the system to begin working as intended.
Posted: Sat Sep 02, 2006 6:51 am
by tmc
Terralthra wrote:QUOTE (Terralthra @ Sep 2 2006, 02:44 AM) They would only be able to stack for a couple games each. I went through it in a thread in gameplay, but basically it would not take more than 10-12 games for each player (running in parallel, naturally) before the gains for the more skilled players and the losses for the less skilled players prevent them from stacking and allow the system to begin working as intended.
Exactly. So regardless of whether we have an autobalance button or not, the system should converge. Autobalancing would not prevent a stack, since it doesnt know whos good at that point.
Posted: Sat Sep 02, 2006 8:08 am
by Tigereye
Tmc...
Can you define 'converge' for me? I'm wondering if we've been talking about the wrong thing all along... it might explain some of the disputes people are having in this thread.
--TE
Posted: Sat Sep 02, 2006 8:16 am
by tmc
converge - as time goes to infinity, the rate of change of people's ELO values goes to zero.
Posted: Sat Sep 02, 2006 8:20 am
by CronoDroid
Personally, I think this reset was a long time coming.
Well done Dev Team. /smile.gif" style="vertical-align:middle" emoid=":)" border="0" alt="smile.gif" />
Posted: Sat Sep 02, 2006 9:08 am
by Terralthra
tmc wrote:QUOTE (tmc @ Sep 2 2006, 06:16 PM) converge - as time goes to infinity, the rate of change of people's ELO values goes to zero.
The problem is that convergence is not the same as accuracy.
Posted: Sat Sep 02, 2006 9:22 am
by tmc
Indeed, ELO might converge but be completely wrong (for example, making all scores permanently constant at 1500 would achieve this). This is a seperate issue. When talking about convergence, we assume that ELO, well, actually works. Does it really? Hard to say. The only real way to find out is to wait a few months and check the rankings.
Posted: Sat Sep 02, 2006 1:22 pm
by Falnyx
tmc wrote:QUOTE (tmc @ Sep 2 2006, 03:16 AM) converge - as time goes to infinity, the rate of change of people's ELO values goes to zero.
Or perhaps better stated, the rate of change of people's elo values go to some small, non-zero number, accounting for the fact that most people are going to improve at least gradually over time. /smile.gif" style="vertical-align:middle" emoid=":)" border="0" alt="smile.gif" />
Posted: Sat Sep 02, 2006 6:49 pm
by zarcain
well, I went from an 8 to a 21. I am definately not that good.
Posted: Sat Sep 02, 2006 7:36 pm
by jgbaxter
Hey tmc, speaking of converging, how do you feel about weighted k-factors, such that a novice might adjust 100% of elo, inter 75%, vet 50%, exp 25%... the fluctuation also is reduced, this is a standard practice for most elo systems...