Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Mate in 1 - but Fritz 6 needs 1 hour!!!

Author: Uri Blass

Date: 11:37:12 08/11/00

Go up one level in this thread


On August 11, 2000 at 14:25:07, Adrien Regimbald wrote:

>Hello,
>
>>Fritz is a chess engine.  This is a chess position.  Fritz can't do it despite
>>the fact that a 5 year old child could easily see the mate in a few seconds.
>>This means that Fritz has failed.  The problem is not that Fritz can't do _this_
>>position, but rather that Fritz spends over an hour on a problem which should
>>take it a tiny fraction of a second.  It shows that the algorithm is doing
>>something wrong.  Imagine a ramp down in problem complexity with an associated
>>ramp-down (not necessarily linear) of time required on average.  It's still
>>going to be taking far too long to solve other pretty obvious problems which
>>aren't as contrived as this one.
>
>
>Fritz didn't fail.  The move which it would have played in normal time controls
>would still have won.  How can one possibly conceive of this as a failure?
>
>Also - are you going to worry about a position which wouldn't happen in a
>million years between any two players with an IQ greater than that of a monkey?
>I can tell you that I won't spend any sleepless nights over such positions!
>
>
>>Checking for mates before you do any searching is not only fast, but it vastly
>>improves search times on positions like this.  I've often seen ColChess analyse
>>4 or 5 ply deeper in some positions than, say, Crafty in the same time.  That
>>was because there are lots of CMs flying around to avoid.
>
>
>As someone involved in computer chess, you should realize that it is a constant
>battle of compromises.  If you make seeing short mates immediately a priority at
>the cost of overall playing strength, and the program starts playing at a
>grossly worse strength overall, would you be happy?  Personally, I'd rather have
>a program that played well in a greater percentage of positions, perhaps at a
>cost to strength of around 200-300 max points (to me the difference between
>getting smacked around by a 2300 and a 2600 really doesn't matter), but I can
>assure you that even this compromise is not on the mind of the people at
>ChessBase who are trying to make the strongest possible chess engine.  I also
>think that the modifications necessary to make a program play correcty in a
>retarded position like this would come at a significant cost to overall ELO,
>which I think is completely unacceptable considering the uselessness of this
>position.

I disagree.

Junior and hiarcs can find the mate in less than 1 second.
Many programs can do it.

Fritz's programmer does not need to do Fritz weaker in practical games in order
to solve the problem.

It only needs to limit the q-search if you search too many nodes.
In most of the cases you do not search too many nodes in the first ply so there
is no problem.

Uri



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.