Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Stupid + super fast = dead !!

Author: Robert Hyatt

Date: 20:34:21 10/07/02

Go up one level in this thread


On October 06, 2002 at 11:46:18, Uri Blass wrote:

>On October 06, 2002 at 11:27:43, Chessfun wrote:
>
>>On October 06, 2002 at 11:05:20, ERIQ wrote:
>>
>>>I just had to post this warning to all chess programers and computer account
>>>managers. Having a computer try to play instantly just to win on time is a bad
>>>idea. Here is a game that a 2400+ program lost to me but used only about 5 secs
>>>the whole game !! and It was a five minute game. Even if it won I would think it
>>>in poor taste to just try to win on time :(
>>
>>
>>Dingbat has a target rating of 2300, in order to get that he speeds up the
>>computer move time, so as to reduce it's ply depth. IOW he isn't trying to win
>>on time.
>>
>>Sarah.
>
>I do not know what is his target but the only logical target of reducing the
>time is in order to reduce the rating.
>
>I am surprised that people think that it is done to win(it is done to win only
>if it is done only when the opponent is in big time trouble).
>
>I also have a parameter that tells the public movei to play faster so if
>opponents think that it is too strong for them they can change the parameter
>from 1 to 60 and movei is going to play 60 times faster.
>
>I never thought that it is going to give movei better rating but I also thought
>about comp-comp and not about comp-human.
>
>The main reason for this function is the fact that winboard does not support
>unequal time control and if I want to test if my program is 5 times better or 10
>times better than another program I have no way to do it without a special
>function.
>
>I found that public movei is more than 5 times better than tscp only when the
>time control is slow enough.
>
>Unfortunately other engines do not support this function so I cannot find how
>much speed movei need to be better than Ruffian but people can get an estimate
>if they play games of x minutes/y moves when y is a big number of some thousands
>thanks to the bad time management of ruffian.
>
>Again I suspect that the difference between the public movei and Ruffian becomes
>bigger when the time control is slower but today I do not care about the public
>movei.
>
>Uri


Actually, moving faster vs humans is not a bad idea, in blitz games.  It speeds
up the tempo,
and the human will generally fall right into the new faster tempo, and explode
tactically.

I found that Cray Blitz did much better at faster speeds.  Everybody _used_ to
play
60 moves in 5 minutes of cpu time against humans, and if the human lasted 60
moves
the computer automatically lost.  This avoided the "operator time" penalty.

I changed this to  go 3 seconds per move until 1 minute had been used on the
real chess
clock, then 2 seconds per move for the next minute, and 1 second per move for
the rest of
the game.  I don't recall it losing more than 1 or 2 games out of at least a
hundred games,
against IM/GM players.  All with me typing the moves in and making the moves on
the
real board and hitting the clock...

IM Mike Valvo said "this is a different game like this, at game/5min, I have a
lot of time to
think while it is searching, so I never get into time trouble.  At this speed, I
have to use my
own time and it is easy to get into an up-tempo game and blunder, or use too
much time
and blunder when time gets short."




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.