Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Junior - Crafty NPS Challenge - a user experiment

Author: Sune Fischer

Date: 03:36:46 11/25/03

Go up one level in this thread


On November 25, 2003 at 05:23:31, Rémi Coulom wrote:

>On November 24, 2003 at 18:38:06, Sune Fischer wrote:
>
>>
>>While reading it I managed to convince myself you are right.
>>
>>Intuitively it seems obvious that if the normalization constraint
>>p0+p0.5+p1=1 is the only correlation between p0 and p1, then
>>p0.5 isn't going to have any say in whether p0>p1 or not, right?
>
>Right, but it is not that simple. What we get from the match is not p0, p0.5 and
>p1, but the number of wins, losses, and draws. p0, p0.5, and p1 are unknown. Of
>course, if you know p0 and p1, you don't need p0.5. That does not oviously
>implies that if you know n0 and n1, you don't need n0.5.

That's true, in statistics everything is done 'backwards' :)

>>The only exception is if p0.5=1, but then we wouldn't have a three parameter
>>distribution in the first place which is sort of the assumption.
>
>This is not really an exception. All the maths work the same if p0.5=1. In this
>case, what you'll get is a match with only draws, so the formula says
>P(p1>p0)=0.5.

Right, unless you played an infinite number of games which all ended drawn.
Then P(p1=p0)=1 which would answer our question.

I guess I'm always assuming an infinite number of games, since I know the
distribution :)

-S.


>>-S.
>>



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.