Ingo,

I actually built a simulator for this once.   I could assign ELO ratings to
the players and they would win or lose based on their probability of winning
as predicted by the ELO system.     It was indeed a good way to get a feel
for this.     I tried things such as random pairings (not taking into
consideration their ratings)  and comparing this to proper pairings.    I
would run the same tournament 10,000 times to see how often the "best"
player actually won.     This was not a formal proof but it was certainly a
good empirical proof that it's important to pair properly.

If I can find that code I will present some data.     The ELO ratings were
assigned in the shape of the bell curve to make it more realistic.   My
formula was something like (random(3000) + random(3000) + random(3000) ) / 3
 and 1500 will tend to be the average rating and there will be relatively
few player a lot stronger or a lot weaker.     That is at least more
realistic that just setting the ratings randomly.

Don



On Mon, May 9, 2011 at 7:33 AM, "Ingo Althöfer" <[email protected]>wrote:

> Hello Nick,
>
> > You are proposing that the tournament should start by pairing strong
> > players with weak players, and claiming that this is more likely to
> > result in the strongest player winning the tournament.  I don't see it.
>
> A formal proof for the general case. "Only" special cases (for instance
> with one top player, (n-1) semi-strong players, n weak players) have
> been proven.
>
> However, one can run "some" simulations to get a feeling for it.
> The following model (with wins and losses only) will do:
> For each player i a positive number r[i] is given, his rating.
> When i plays j, i will win with prob
> r[i] / ( r[i] + r[j] ),
> and j will win with
> r[j] / ( r[i] + r[j] ).
> For simplicity we assume that different games run independently of
> each other.
>
> Select some set of 2*n strengths and try
> different pairing schemes to see, how often
> which player wins.
>
> Ingo.
> --
> Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
> belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de
> _______________________________________________
> Computer-go mailing list
> [email protected]
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to