Author: Tord Romstad
Date: 23:58:26 01/18/04
Go up one level in this thread
On January 18, 2004 at 11:44:56, Russell Reagan wrote: >On January 18, 2004 at 06:05:06, Tord Romstad wrote: > >>Has anybody else experimented with similar ideas? Is it possible to make it >>work? > >Gian-Carlo used 'proof number squared' search in his program. I don't know if it >is part of the normal search, or if it is a seperate thing he implemented just >to play with. My guess is that Gian-Carlo originally implemented this for use in suicide chess (which old versions of Sjeng were able to play), but also makes use of it for mate searches in normal chess. >http://chessprogramming.org/cccsearch/ccc.php?find_thread=312368 In that thread, GCP mentions that PN2 helps him to find a particular forced mate quickly. This is not very remarkable, it is well known that PN (and related algorithms like PN2, PN*, and PDS) excel at such problems. My idea was to use a PN-like algorithm in a more general way, by formulating some sort of precisely defined goal and use PN to decide whether it can be achieved. I think this can be particulary effective in the endgame, where the result often depends only on whether one of the players can achieve some particular goal. My long-term goal is to have a highly selective main search with a very big, slow and sophisticated evaluation function, and to include several fast and simple special-purpose "sub-engines" to assist the main search in tasks like tactical verification searches, threat detection and move selection. I've already taken the first tiny step in this direction. In my latest development version, at nodes where the static eval returns a so clearly winning score that only a direct tactical refutation could prevent a fail high, I do a null-move search with a simplified evaluation function. It is too early to say whether this simple idea works, but my initial test results have been promising. 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.