Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: KB*KP* -> some first promising results

Author: Tony Werten

Date: 01:00:46 02/04/05

Go up one level in this thread


On February 03, 2005 at 09:42:29, Thomas Mayer wrote:

>Hi Tony,
>
>>>> I have coded a few others (KBP[and more Ps] vs K with all rook pawns, and
>>>> KPPPP... vs K all rook pawns), but don't use it at the moment.
>
>>> They are part of my KP*K and KB*P* recognizers... this is quite simple and
>>> almost cost zero. If all pawns are on a rook file and the most advanced
>>> pawn is a draw in the bitbase then the hole position is draw...
>
>> I wonder. Did you check the cases where it is a draw because the king
>> captures the first pawn ?
>
>> [D] K7/8/PB6/Pk6/8/8/8/8 b
>
>in that case the recognizer returns a failure because the weaker side can
>capture a pawn.

OK so you covered that. You don't have to return complete failure. In XiniX I
also return AT_MOST_DRAW and AT_LEAST_DRAW flags . With 2 flags in a row you can
return 0 very often with out/limited search.

The flags do leave a philosophical problem. Suppose I have 2 positions wich
normally evaluate to the same positive value.

My recognizer says the first 1 is at least a draw, the 2nd isn't recognized.

Should they return different scores ?

One can say: Fist is a garantied draw, score it higher.

One can also say: 2nd isn't recognized, so a bigger change it's not a draw.

>So it see's the draw a bit later -> IMO the task of the

It shouldn't. It's a win for white :)

Tony

>recognizer is
>
>a) to be fast
>b) to return only CORRECT results or tendencies
>
>if the return of the recognizer is totally wrong you have a problem because it
>will hide the result in the tree...
>
>e.g. one position I had troubles with:
>
>[D] 5k2/4p2p/6P1/3K4/8/4B3/8/8 w - - 0 1
>
>The solution is 1. Bh3+ with
>1. ... Kg8 2. g7 Kf7 3. g8Q+!! Kxg8 4. Ke6 Kh8 5. Kf7 e6 6. Bg7++
>
>It was not easy to find rules to return a failure in the KB*KP* recognizer in
>all positions... Now it works fine... :) You see that I test this with studies
>currently...
>
>Greets, Thomas
>
>P.S.: And I have removed lots a bugs yesterday -> I feared that it would work
>worser -> but it gets even better and faster...
>PPS.: Hm, rethinking your position I think instead of a failure I could at least
>return a 0 upperbound for white... cool cool... :) I have some work... :)



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.