Author: Christophe Theron
Date: 11:22:58 12/11/00
Go up one level in this thread
On December 11, 2000 at 08:17:10, Thorsten Czub wrote:
>On December 11, 2000 at 07:31:11, Enrique Irazoqui wrote:
>>Cherry picking... Analyze also the games that Fritz won and Gambit lost, and
>>already posted wrong attacking evals of Gambit when overestimating a king
>>attack. Now and then even the strongest programs evaluate positions wrongly,
>>meaning that chess has not been solved, which might come as a shocking surprise
>>to you. :)
>
>you are completely on the wrong track. gambit tiger (as cstal) was designed
>to build strong king attacks. its a feature.
>of course gambit-tiger will overestimate many king attacks. this is because it
>is as inaccurate enough.
>but even beeing as inaccurate as it is: it is still good enough for getting 50%
>against fritz.
>
>i was talking about a thing, that is in fritz for versions.
Thorsten,
You are bringing the subject up again, so I have to repeat what I have said
yesterday.
It's not a "thing" in Fritz.
It's probably a pragmatic choice of the programmer.
The choice is to pay little attention to king attacks. The reason is that:
1) the opponent has to be damn strong to build and conclude a king attack
without Fritz noticing the trap with its search.
2) many king attacks are just not correct and will fail against a good (not even
excellent) defence.
3) There are very few chess programs that will consciently try to build an
overwhelming king attack, so ignoring them does not hurt for the SSDF result.
The choice is dubious given that the goal in chess is to mate the opponent
(presumably by first building a strong attack against it), but if it works then
it's probably not so bad.
So, IMO, Fritz gives a little weight to the "king attack" part of its
evaluation.
Generally, a side building a king attack has to make some positional compromises
in order to achieve it. If Fritz values these compromises higher than the "king
attack" terms, then it considers that the side which is attacking has a weaker
position (for example, the attacker can weaken its pawn structure in order to
open lines to the opponent's king, and if the attack fails then the weak pawn
structure will later be a terrible handicap).
It sounds strange to a chess player, but given the idea explained above, it is
logical. And it's not so stupid, or else Fritz would be an obviously weak
program, which is not the case.
Now what happens when Fritz opponent's makes a lot of compromises in order to
build its king attack? Fritz believes the opponent has got an inferior position.
So it believes that the opponent will take a draw by repetition if it gets a
chance to.
There is an interesting property of strong king attacks: if the attack is good
enough, as I have explained yesterday, the search will first see that the
attacker can force a draw by repetition (generally a sequence of checks). If you
let the search go much deeper, it will see that the attacker can not only force
a draw, but also chase the king until it is mated. But it takes several
additional plies to see it, and much more time.
So in the case the opponent really builds an overwhelming king attack, Fritz is
going to believe for a while that the opponent will force the draw. Later, and
only later it will see by search that the opponent can not only force a draw but
also mate the king.
That's all. Fritz is not the only program to have this "problem". Tiger 12
probably has the same problem, because my king attack evaluation was not very
good at that time. Chess Tiger 13.0 and Gambit Tiger 1.0 are different because
they have a lot of knowledge about king attacks.
I don't want to discuss the choice of the programmer to pay little attention to
king attacks. This is not my point. My point is that the explanation of the
"draw score" behaviour is very simple, and is not such an intrinsic weakness of
Fritz.
>if you call this cherry picking or
>
>"Now and then even the strongest programs evaluate positions wrongly,
>meaning that chess has not been solved, which might come as a shocking surprise
>to you. :)"
>
>is only desinformation of the people.
>
>what you say is on a low level.
>it is so general that it isn't an acceptable answer.
>
>i said: one reason why fritz loses is, that it is blind.
>when it is blind, it has these draw scores.
It is not blind, it has chosen to pay little attention to king attacks, and it
is often right.
It's more about philosophy than about blindness.
>we now see this behaviour is in any version of fritz.
>
>and your only answer to this ( i gave concrete examples ! even examples
>of your OWN data, i am sure you had these data but never understood the data you
>made yourself ! seems you have the same "bug" fritz has ! 0.00 Enrique ?
>well : 4.2 for moritz, and 0.00 for enrique !)
>is:
>
>chess is not solved thorsten !
>
>
>stay on this level. it makes you look very good enrique.
Why are you attacking Enrique?
Christophe
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.