Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty time handling problem?

Author: Robert Hyatt

Date: 07:49:05 12/21/01

Go up one level in this thread


On December 21, 2001 at 05:49:43, David Rasmussen wrote:

>On December 21, 2001 at 00:11:31, Robert Hyatt wrote:
>
>>On December 20, 2001 at 17:57:35, David Rasmussen wrote:
>>
>>>On December 20, 2001 at 14:44:18, Robert Hyatt wrote:
>>>
>>>>
>>>>Because it _assumes_ it will save time here and there by "pondering" correctly
>>>>and it adjusts the time target for each move based on this assumption.  But
>>>>the assumption is _wrong_ since it isn't pondering at all.  The time allocation
>>>>for ponder=off simply needs more tuning.  But I don't _ever_ play serious games
>>>>with ponder=off so I never test or tune for that...
>>>
>>>I can understand that you get bad time allocation, without pondering, that is,
>>>too much time is used on a move, or not enough, but to actually lose on time, is
>>>another matter. But I see your point.
>>>
>>>/David
>>
>>
>>I can't see how crafty can _ever_ lose on time.  At least under an O/S with
>
>No, that was what I was saying. But that was what Jouni Uski was seeing. And you
>seemed to say "with ponder off, it _can_ lose on time", and now you're saying
>(like me) "even with ponder off, it cannot lose on time". I'm confused.

OK.. let me be a bit clearer.  I time to the nearest .01 second.  Which means
that after crafty gets down to one second left on the clock, it is going to
take at least .02 seconds per move (I don't stop until I use _more_ than the
target time to help with those cases where the target time comes out to "zero"
in rare cases.  After 50 moves, it _must_ lose on time.  But ponder=off could
cause it to get to such a "low time remaining" situation in the first place.

So "ponder=off" would be a contributing factor as it would use too much time
too early in the game, and then suffer later...




>
>>
>>My "stress test" for unix is to play "game in 1 second" games.  (you have to
>>hack xboard to support this, or else you can try game in 1 minute which is also
>>a tough test).
>
>Why do you have to hack xboard? Can't you just make an incremented game with 0
>increment and 0:01 minute ? Works for me.


Last time I did this it started the game with 10 seconds on the clock.  I
simply hacked an old version of xboard and saved it for this "stress testing"
stuff.  If newer ones don't have that problem, I will give them a try.  I
also think that my older versions only accepted time in units of one minute,
not one second, unless I missed something.




>
>/David



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.