Author: Ilya P. Kozachenko
Date: 20:58:25 08/03/98
On November 09, 1997 at 11:31:49, Alvaro Polo wrote: >I wonder if a chess program could be made which used two different >strategies in parallel (using two processors). > >On the one processor it would run a very knowledge based algorithm, >something like CSTal appears to be. > >On the other processor it would run a fast and deep searcher. The >tactical lines would be found by this second algorithm and forwarded to >the knowledge based one signaling them as lines to avoid. On November 09, 1997 at 14:38:40 Robert Hyatt wrote: >this has been done. See "Phoenix" by Jonathan Schaeffer. He used two >parallel search engines, one a full-blown search+eval, the second an >aggressive null-move search with material-only, which could search a >couple of plies deeper than the other. It worked, but only "so-so" >because >the fast search can find tactical things, but not positional things. So >it >could find a way to win a pawn, but wreck the position in the process. >Or >find a way to defend the pawn, but wreck the position. It was hard to >"coordinate" the two searches to decide which is correct... And what about next idea?: Since hard coordinating let's use 2 different algorithms not in parallel. Program could define whether position is "more tactical" or "more positional" and apply corresponding algorithm. We would obtain a program with 2 kinds of play available, which changes its style during the game (like DeepBlue :) and isn't worse than good "knowledge-based" or "speed-based" program alone. Any suggestions, please. WBW, Ilya.
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.