Author: Alessio Iacovoni
Date: 15:24:46 10/05/98
Go up one level in this thread
On October 05, 1998 at 11:43:23, Komputer Korner wrote: >On October 04, 1998 at 05:27:00, Alessio Iacovoni wrote: > >>Why don't chess programs automatically implement a random chess feature changing >>(by some say .05) the evaluation algorithms? It would give a much more lively >>chess play. What happens, otherwise, is that the chess program will, at same >>time controls, play the same move over and over again, making the game boring. >>Also.. it would prevent other chess computers from adapting their strength to >>that particular chess program. I believe it should be quite easy to implement... >>especially on some specific parameters : i.e. passed pawn. Others could be left >>the same in order not to weaken the program more than much: i.e. king safety. >>These parameters can already be manually modified in many chess programs.. but >>there is no feature (that i know of) that does it automatically in a "quasi" >>random way. >> >>I don't know much of chess programming, but in phylosophy "theory of chaos" has >>helped and is helping more and more.. some principles of chaos have been applied >>to speech recognition for example (aristotelian).. It may seem a paradox but.. >>who says that the best move is going to be the best after all? > >The problem with this is when putting in moves to the hash table. Constantly >changing the evaluation algorithm will result in moves being undervalued or >overvalued compared to the algorithm being used at that node unless you clear >the hash table at every node. And with alpha beta only having the score of the >best move found so far, you won't know whether the 2nd best move is close enough >to be able to consider it as an alternative. Preprocessors like Nimzo decide at >the root what the algorithm will be for that node but the problem with that is >that they are depending on a 1 ply search to adjust the parameters of the >search. These may or may not be relevant the deeper you go into the search. >Since Nimzo 99 is so strong, it must have solved this problem by modifying it's >preprocessing in some way. >-- >Komputer Korner Actually I wasn't thinking of changing the evaluation algorithim at all but just randomly modifiying the weights given to some specific chess features such as passed pawns, pawn structure, king safety and so forth by a very small percentage in order to favor some degree of randomness in the style of play. Fritz for example already allows to play in "handicap" mode. Why not do that automatically (with very small changes) before each game, also tournament ones. It could held avoid other programs from exploiting some specific mistakes or weaknesses of the program in games previously played.
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.