Author: Dave Gomboc
Date: 22:02:30 09/28/98
Go up one level in this thread
On September 28, 1998 at 02:08:58, Tom Kerrigan wrote: >It seems to me that if you have lots of "modules" to handle the types of >positions you might encounter, the modules will all be very similar to each >other (e.g., almost all of them will have a doubled pawn term), and it would be >inconvienant to switch modules during a search. If you have one huge evaluation >function that turns terms on and off depending on the position, it seems like it >would do better and take less effort to write. > >-Tom > >On September 27, 1998 at 15:14:38, Fernando Villegas wrote: > >>On September 27, 1998 at 15:02:24, Tom Kerrigan wrote: >> >>>This sounds all fine and good, but it turns out that most programs already do >>>this to some degree. Every term in an evaluation function isn't evaluated in >>>every position. Many terms depend on particular circumstances or the amount of >>>material on the board, so while the same function is being executed all the >>>time, only certain parts of it are used and with "appropriate" values. >>> >>> >> >> >>Hi Tom: >>I imagined it, but the question is how much is done in that way. The "some >>degree" is the question. In fact how much of a superior intelect the first >>module is. Maybe just a table to say "there are few pieces, this is an ending". >>But something more intelligent to smell where the game is going? >>Fernando >> >>-Tom >>> >>>On September 27, 1998 at 14:32:27, Fernando Villegas wrote: >>> >>>> >>>>Reading a post by Schroeder where he explain how his anti GM works only if >>>>certain circumstances happens, I wonder if the same principle of a program that >>>>is in pieces or modules differently activated according to circumstances has >>>>been made before, and at which degree. I wrote a long and cumbersome post long >>>>ago about that and I do not remember if a programmer answered it or not, so with >>>>your pardon let me give of that idea a very tightened resume. >>>>My idea was: lets a program be constituted by a module dedicated only to >>>>determinate what must be searched and then, with it, a normal program cut in >>>>pieces differently assembled together according to occasion. The idea was to >>>>simulate what a good player in fact do. A good player does not calculate every >>>>available move applying to each of them his entire set of knowledge; he, first, >>>>decides what to look at. There is some smell of tactics in his king side? Should >>>>he look for them? Instead, should he thinks in terms of strategic advantages to >>>>get in the queen side? etc...Once he has grasped the essence of the position, >>>>that is, where action is, then and only then he begins to use that specific part >>>>of his knowledge acumen necessary for the task. >>>>Why not a program with a kind of device similar to that to decide where the >>>>action is and then use the necessary modules for the rest of the job? By >>>>example, if the position calls for tactics, then he would uses only the >>>>algorithms oriented to tactics, maybe with full width searching, etc. But then >>>>if an strategic approach is necessary in the queen side, he drops the tactic >>>>part of the engine and concentrates all his speed in evaluating strategic >>>>parameters such as pwan races, etc. >>>>Of course some mix should be made: maybe after strategic consideration a quick >>>>look for eventual tactics would be ever necessary to avoid mishaps. But I do not >>>>want to enter in details about this because i suspect that this has already >>>>invented or rejected. Could a programmer tell me about this? Did I pick a good >>>>idea or just I picked an unfeasible one? >>>>fernando Well, the blackboard approach is used to attack many AI problems. It should work for assessment of a chess position too. I guess I'd be worried about performance, the methods already in use are so darned fast compared to a casual blackboard implementation. :-) Dave Gomboc
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.