Re: computer-go: Re: Generalized Ranking System

["Mark Boon" <tesuji@xxxxxxxxx>]

The assumption may be wrong, but that is not shown by the statistics. The
statistics show the winning chance against players with a higher dan/kyu
rank. For kyu players this seems almost linear, but the higher one gets up
the dan-ranks this breaks down. This is natural because as players get
stronger their winning percentage improves when they get a stone handicap.
This is because stronger players make smaller and fewer mistakes and it
becomes increasingly difficult (or maybe I should say 'unlikely') to beat
them with a stone handicap. This non-linearity makes it hard to map dan/kyu
ranks to an ELO rank. But the statistics don't really show easily that the
ranks are not associative.

In general in Go associativity in ranks is also assumed. When player A can
beat player B with one stone handicap and player B can beat player C with
one stone, we usually assume player A can beat player C with two stones.

The data in the statistics are interesting in itself that they seem to
indicate less difference between players of one rank difference than what I
had before.

>
> As for the discussion about God:
>
> If we don't believe in freedom of expression for people we despise,
> we don't believe in it at all -- Noam Chomsky
>

This is a good quote, I totally agree.

    Mark

----- Original Message -----
From: Cuijpers, P.J.L. <P.J.L.Cuijpers@xxxxxxxxxxxxxxxxx>
To: 'Computer Go' <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Wednesday, September 27, 2000 5:46 PM
Subject: computer-go: Re: Generalized Ranking System


> I have some criticism on Mark Boons post:
>
> He wrote
>
> > I used a simple notion that player A
> > is one rank higher than player B if he wins two out of three games. I
> > also assume that if player A is one rank higher than player B and player
B
> is
> > one rank higher than player C, player A is two ranks higher than player
C.
>
> > (This is not a trivial assumption, but the discussion gets too
complicated
>
> > otherwise.)
>
> This assumption turns out to be wrong.
> Maybe due to the handicap system or maybe due to human factors we can
easily
> verify at:
>
> http://www.european-go.org/rating/statev.html
>
> that a 20k player has a chance of 36% of winning against a 19k, 35%
against
> a 18k and 24% against a 17k while a 3D player has a chance of 36% against
a
> 4D, 20% against a 5D, 8% against a 6D and no chance against a 7D player.
> The mapping of the kuy-dan to generalized ranking is therefore strongly
> non-linear. Pity.
>
> As for the discussion about God:
>
> If we don't believe in freedom of expression for people we despise,
> we don't believe in it at all -- Noam Chomsky
>