Author: David Dory
Date: 04:15:12 03/10/03
Go up one level in this thread
On March 10, 2003 at 01:24:01, Pham Hong Nguyen wrote: >On March 10, 2003 at 00:05:06, K. Burcham wrote: > >> >> >>When program A is playing against program B in a game, we know that program >>A is looking for best move to play based on its own search for next move to >>reply to its own move. Then searches for next best reply to that move, and next >>best reply to that move, etc. All this based on its own search. >>In other words, if a program is playing white, it is also playing black at the >>same time, in its own eval. A programs eval doesnt even know its actually >>playing against another program, it thinks it is playing against itself. >> >>I was wondering, lets say this game is being played at one of the chess sevrers. >> >>1. opponent A has two computers. On one computer opponent A has program x. On >>his other computer opponent A has program xx. Both with very strong hardware. >> >>2. opponent B has one computer. On his computer opponent B is running program >>xx. Very strong hardware. >> >>what if opponent A was able to enter expected moves of program xx into program >>x. It would seem that this would be advantage. > <snip> I believe it's always a little daft to describe what the computer is "thinking" - just because it isn't "thinking" at all. Nevertheless, sure it would be a help if any chess program could get it's opponent's next move, in advance of it actually playing the move. As the opponent's program continuted to evolve, any program so tuned would become obsolete, giving wrong moves, and thus become a hindrance instead of a help to your cc program. This sounds a lot like the advantage of networking some computers together to play against one opponent. One advantage the networked comps would have is each computer could guess a different next move by the opponent, and ponder a good reply to it, in parallel. With enough computers networked together, all replies could be pondered, even before the move was made by the opponent, on every move. Since a lot of time might be saved on the clock, an improvement in ELO seems certain. How much improvement would depend on your program's ability to select the opponent's move for pondering, currently. If you only are getting 55% correct selection for pondering, OK, that should give you an extra 1 - 3 ply depending on the position, the hardware, and the time controls of the game. If your program is selecting the correct move of the opponent 85% - 95% percent of the moves, obviously the increase in performance would be negligible. If your idea included using CPU cycles in the networked computers for working on searching and evaluating the various possible moves in a coordinated manner, (like a dual or quad except over a network with more computers), that would be hard to program, and suffer hugely from the (relative) slowness of the network. The former idea would be easy to implement, but networking many computers together to do a coordinated search through the move tree would be a tough job and probably a waste of time. Dave
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.