Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: how to evaluate KRP vs KR

Author: Sune Fischer

Date: 15:08:29 06/05/04

Go up one level in this thread


On June 05, 2004 at 16:53:20, Tord Romstad wrote:

>On June 05, 2004 at 15:26:53, Anthony Cozzie wrote:
>
>>We need to get Nalimov to write some W/L/D tables.
>
>I am not sure I agree.  Implementing simple high-level rules for evaluating
>endgames
>like KRPKR correctly is a difficult, but very valuable excercise.   The reason
>is that you
>will sometimes discover principles which are also useful in more complicated
>endgames, but which are much more easily noticed when there are just a few piece
>left.

The problem is often that if you add a piece it's a lot more unclear.
Eg. compare BP-R to BPP-R or BP-RP.

The first one is draw with a very high probability almost independently on the
specific position, but the other two are very unclear and requires a more
careful analysis to even make a guess.

>  If you rely too heavily on bitbases and tablebases in the early phases of
>development, you lose the chance to make such discoveries.

True.

>This applies to human chess players as well as computers.  I spent a lot of time
>studying the KRPKR endgame when I was young.  I cannot remember a single
>tournament game I played in which this endgame appeared on the board, but
>the heuristic knowledge learned by studying this endgame helped me save a
>lot of half points in rook endgames with numerous pawns.  Studying basic
>endgames gives you a unique chance to learn about the strengths and
>weaknesses of the individual pieces and how they interact.

The KRP-KR endgame happens surprisingly often so it is worth having special code
for that, IMO.

It's also rather easy to get good estimates on w/d/l probabilities. :)

>My experience as a tournament player was that knowledge of basic endgames
>decided a much bigger fraction of the games than concrete opening knowledge.
>This is one of the reasons I find it hard to understand the FRC enthusiasts.  In
>the endgame, of course, FRC and classical chess are identical.  If you want to
>reduce the importance of preparation and knowledge in chess, you should
>invent a variant where the endgame is radically different, not a variant where
>the opening setup is random.

Are you spreading the FRC discussion to new threads now? :)

I think conceptually there is a rather big difference between endgames and the
opening.

The opening is a specific position which can be studied deeply, while endgames
tend to be more about heuristics.
There are a few important endgames of course, but if you look at the volume of
litterature on openings compared to endgames, you will probably find the endgame
to be much less heavy on theory, at least that's my impression.

Another trend in endgames is a very small number of pieces. I think it won't
help much to add a few pieces in the beginning of the game, the endgame will
still be the endgame.

>>Preferably in C rather than
>>C++, considering that 2/3 of all instructions in Zappa are from tbindex.cpp :)
>
>Here, however, I agree 100%.  :-)

It seems something isn't being generated optimally compared to similar C code.
I think it would be better to improve the C++ compiler instead of rewriting
everything to C.

-S.
>Tord



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.