Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: What is Prescan Heuristics ?? NT

Author: Vincent Diepeveen

Date: 10:46:04 10/13/00

Go up one level in this thread


On October 13, 2000 at 11:50:27, Kim Roper Jensen wrote:

>No text

it's pretending you know in advance what is the best move and just
let the program figure out the tactics. it's fast and of course my
first line is a rude outline of what it is, but it's definitely no
compare to a leaf evaluation.



So the 2 things against each other:
   - prescan heuristics/ piece square tables:

fill a table BEFORE you search and apply
there the knowledge

so for example for each piece at each square
and in your makemove only add the score from the tables and do nothing
in your leafs (don't see why one would need a qsearch in fact for
such programs a 'guess' function should be plenty enough).

No evaluation needed. Even passers you can easily do incremental,
no problem.

  - leaf eval: you can sure use the makemove to build up datastructures,
    but the evaluation is NOT dependant upon the moves made, but only
    upon the position and you go scan the board position in each leaf
    for chess knowledge.

So the big difference is for example when exchanging queen: chess
tiger used (don't know whether it's still a preprocessor)
to be happy before queen exchange, then after queen exchange it's
suddenly very unhappy. Old fritz versions except the latest 6a,6b etcetera
were also preprocessors.

Of course many programmers figured out that using a preprocessor only
is very hard to play chess, so usually a few big scores they do in
eval like passed pawns. Many old programs simply had no king safety
at all.

Nimzo is a classical case of preprocessing too. if you checkout its book
you'll see it hardly plays openings with very tough pawnstructures
without having all tough lines in book, as otherwise the preprocessor
makes big mistakes as it doesn't see the position change in the root
of course.

Now where is the big advantage of preprocessors apart that you can do
it faster, why have they been so pretty successfull in the past?

I think very important reason is that it's also very easy to debug a
preprocessor when comparing that with a leaf evaluation.

Greetings,
Vincent




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.