Author: Cliff Sears
Date: 02:45:10 08/11/04
Go up one level in this thread
On August 10, 2004 at 15:45:18, Volker Böhm wrote: >Hi, > >sorry that I can´t help you with your question, but perhaps I can give a hint. > >In every position, even opening positions, chess engines are searching full >moves until a certain (flexible) depth and then doing captures/recaptures >(quiescense - search). I suppose you know that but I need it for explaining the >following. > >Because of captures/recaptures a chess engine evaluates different type of >positions (opening, mid-game, endgame) in many poisition (sure it cannot go back >to a mid-game position once in endgame). > >Thus if a chess engine is doing what you are asking for, it must change the >algorithm (the personality) for different nodes of the same search tree. >In some situations it will be very hard to decide which algorithm to take, is it >a midgame poisiton or allready endgame? > >If you have a function that decides if it is opening, midgame or endgame then >the function will suddenly switch just because of a tiny change in the >situration. This is a large drawback for a chess engine because the searchvalue >gets instable reducing the search depth and the quality of search result. > >The only way to avoid this is to evaluate the position to different aspects >multiply those results with a weight and then add the aspect-values for the >total value. > >For example a position is 60% midgame and 40% endgame. Then you could call >eval = eval_midgame * 0.6 + eval_endgame * 4.0 > >Mostly this approach is too expensive to calculate. Better to calculate each >aspect of game once and then add it weighted according the game-state to the >total eval. >This works better for chess. > >Thus I doubt that you will find a good chess engine doing what you asked for. > >Greetings Volker I have tried to use Shredder's Triple Brain function to see if that would be a rough approach, but using one engine known for being more tactical and the other more postional. I could never quite figure out the algorithm used by Shredder to make its decision. It was interesting to follow along with the old Dover book "500 Master Games of Chess" to see how often it would follow the games or the analysis offeed by Tartakower"
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.