Author: Bob Durrett
Date: 06:25:18 12/08/02
Go up one level in this thread
On December 08, 2002 at 00:31:07, scott farrell wrote: >it occurs to me that most programs (well, mine anyway) use the same val function >for both sides, the same king safety etc etc. > >Obviously that alphaBeta assumption is the opposition will play their "best" >move. > >What if you different eval function for the opposing side? > >Is this what personalities are? or fritz etc, claiming to have "anti-GM" code? > >My idea is that the eval function for a human player may be different. Perhaps the main stumbling block would be difficulty in accurately modeling human chess, and coming up with representative moves for the humans. For example: Suppose it was desired to optimize [by settings] the chess engine for play against a 1500 amateur human chess player. For the sake of discussion, assume the chess engine is playing the White side and the human amateur is playing the Black side. Then whenever it was Black to play, you would want the engine to give highest priority to moves likely to be played by the human amateur. But how to decide which moves were typical of human amateurs [at the 1500 level]? Human amateurs [at the 1500 level] make many moves which stronger players would call "blunders." But when the amateurs are not blundering, they may play moderately good moves. Another problem common in such amateur play is that the human amateur OFTEN encounters positions in which they are clueless as to what is going on in the position. The position may be too tactically complex, or it may be of a positional nature. The latter type of position probably looks like meaningless "quiet" positions to the amateur. This is not always true, however. Amateurs have their good days, when they play much better, and their bad days, when they are "blind as a bat." Bob D. >Such that >the search function will try to eventually find a position that highlights the >variance between the 2 eval functions, and theoretically finds that best >anti-human or ant-computer position in favour of your own engine. > >I think just dropping a few eval features for your opponent, or adding a few >extra for yourself, might change play. It might even allow a programs to attempt >to play "traps" on an opponent - which current programs dont play - as if the >program can see the trap, and the solution, boring old alphaBeta wont play the >move. > >Has anyone tried anything like this? > >or am I just warped? > >Scott
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.