Author: Fernando Villegas
Date: 17:29:14 03/11/00
Go up one level in this thread
On March 10, 2000 at 16:36:01, Inmann Werner wrote: >On March 10, 2000 at 15:50:26, Fernando Villegas wrote: > >>Hi InmmaN: >>Why not to implement a duaL engine program? >>Two ways to do it I believe: >>a) First the tactical engine search alone and if get something enough good the >>move is performed without nothing else. If tyhe search does not return nothing >>clearly good, he positional engine is putted to work. So in may cases you does >>not expend more time -only the tactical engine works- and when you do is beause >>you need to do it. >>b) a "supervisor engine" examines the position according to general criteria and >>puts one of the engine to work. This seems to be very tricky, but it is the way >>human GM works. They first look what the position ask -sacrifices, attacks, >>defense, improvement of the position here or there, etc.- and only after that >>they begin to look at specific moves. >>Fernando > >maybe perfect idea for a 2 CPU machine? (parrallel search tactical and >positional) >if you only have one cpu, when to stop the tactical search and beigin the >positional one? maybe one ply deeper there is some tactics, the opponent gets, >and you are dead (compi vs compi). Well, this a matter of practical answer as in everthing else. Of course you can make a mistake, but I suppose you are not trying to write the perfect chess program, but a better one. To me, decision could be something like: 10: if nothing = or > than 0.50 points inside an horizon of 10 plys, then goto 20 20: put off tactical engine and put on positional searcher. The point os not of with that technic you will commit someday mistakes, but if yu in average will get better games. Fernando > >I am quite happy (for amateur prog) with my search algo (=taktic), but my >evaluation terms sucks. > >Today I began to rewrite it... >Example: >First i took the horse.. (middlegame) >a) more in the middle of the board is good. (not at a and h line...) >b) Outpost is good, especially near the opponent king >c) make it active, not stay at 1st line. >d) mobility? I think no.... >e) ????? enough for positional...., when only looking at the isolated horse. > >Now B (Läufer?) >a) mobility >b) not the bad pawn capture at a7 >c) ...... > >and so on. you stuff in and in, especially the rooks are nice and have much >"knowledge", the queen has only less, and the pawns are an own evaluation...., >same "king safety" and then hashing should not be disturbed.....(I like it, if >the same position always gets the same evaluation..) > >And if you put everything in, it should fit together. You can take nothing of >special positions, cause it should work all time. > >At the end you have some code and now you should decide, if it is worth. >But how decide it? Is it now better or awful. >I do not have the money to play thousand of games at ICC or my wife kills me. So >I take test positions..... (holy sh..) >It is not important, that the prog takes the right move, it is more important >that it takes it cause of the right idea. Only with absolutly nontactical >positions, you can then test the eval, or you test your search algo.... > >Got a little long, excuse... > >Werner
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.