Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: I take my Crafty comment back (resp CCT4)

Author: Dann Corbit

Date: 22:18:22 01/28/02

Go up one level in this thread


On January 29, 2002 at 01:09:15, Jouni Uski wrote:

>I didn't know, that it lost 2 games by forfeit! So Crafty's result was quite
>good - it was actually only program not to lose a single game by playing...

I think probably Hiarcs suffered some simlar fate.  Or just had a bad spate of
probability.   The TPR and ELO are miles apart.

Of course, it is not difficult for an excellent program to lose to an inferior
one.  In fact, if you have a program which faces a program 195 ELO points
superior, it will still win (on average) 1/4 of the points.

In fact, let's imagine a program with an ELO of 2300, and it must face 5
programs with an ELO of 2495.  The chances of getting a point from each of them
(this is fuzzy logic, because I am not splitting into draws and wins) is:
(1/4)^5 = 1/1024
Which means that once every 1,024 experiments, that absurd sequence happens
[well, not exactly, but you get the idea].

If (for instance) instead of Junior winning the contest, it had scored only 50%,
I would not have assumed that Junior was slipping.  Rather (based on excellent
results which rely on many times the volume of data) I would assume that it is
[much more likely] an abberration.



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.