Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A test position for chess programs(importance of tablebases)

Author: Uri Blass

Date: 01:40:11 01/20/01

Go up one level in this thread


On January 19, 2001 at 22:25:46, Dieter Buerssner wrote:

>On January 19, 2001 at 19:13:56, Vincent Diepeveen wrote:
>
>>Uri again like all positions of yours,
>>this is NOT a forced combination.
>>
>>It's unforced.
>>
>>All you need is searchdepth, 1M nodes a second.
>>A material evaluation, not even piece square tables
>>needed, and not a single extension.
>
>Vincent, I cannot see at all, how this applies to the position of this
>discussion. I am not sure, if Rc3 draws here, but certainly Rb6 will loose.
>Without TBs, my engine can see the loss of a pawn after Rb6 and after Rc3.
>Nevertheless, with TBs rather soon it is seen, that Rb6 is leading to a forced
>mate, while this is not seen for Rc3. Actually, without TBs, after Rb6 and
>after Rc3 I get essentially the same scores at the same search depth. I have
>no idea, how Rb6 can be avoided based on a material evaluation only.
>
>-- Dieter

I can add that the only moves that maybe do not lose may be Rc3 and Rc2(I
checked with Crafty and it can see a forced mate for white after every other
legal move very fast except Rb6 when it needs a long time).

After Rc3 or Rc2 it can only see less than +1 pawn evaluation for white at depth
20.

Maybe big hash tables and some selective search only for black can help programs
to solve the position after a long search because the number of relevant
positions before the tablebase hits is relatively small.

Usually after a pawn move of white capturing the pawn leads to a draw and I
assume that the program tries capturing the pawn in the qsearch.

The slective search that I suggest is simply not trying to move the black pawn
without capturing and not to increase the distance between  the black king and
the white pawn unless it captures the rook(if black has no choice to do it
consider the position as a loss for black).

If it is possible to prove by this selective search that black can draw by
forcing repetition then it is clearly a draw.

It is interesting to know in how many cases increasing the distance between the
black king and the white pawn without capturing is the only good move for white
in KRP vs KR endgames.

You should consider only cases when black can avoid increasing the distance
without letting white to capture the rook.

I do not imagine other cases when increasing the distance between the black king
and the white pawn is the only move but it is possible that there is something
that I do not think about and this is the reason that it is important to check
all the KRP vs KR positions in order to find exceptions in order to improve the
rules.

I think that this information may help programs to define some selective search
rules for KRP vs KRP endgames(in this case increasing the distance from both
pawns seem to be usually illogical).

I believe that selective search can be productive in endgames but programs
should be careful to select the moves correctly and checking the rules on
tablebases positions can discover holes in the rules.

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.