Author: Dann Corbit
Date: 16:26:21 07/12/05
Go up one level in this thread
On July 11, 2005 at 13:47:13, Uri Blass wrote: >On July 11, 2005 at 13:18:06, Djordje Vidanovic wrote: > >>On July 10, 2005 at 21:38:58, Lin Harper wrote: >> >>> >>> >>> Hi Djordje, >>> In addition to the interesting info in this thread, there are just a couple >>>of things I would like to add. >>> In the case of search depth versus knowledge, you quite rightly pointed out >>>that search depth *is* knowledge. Since the basic problem of a chess programmer >>>is correct allocation of limited processing capacity, perhaps in the case of >>>Fruit2.1, the programmer has succeeded in his effort to include only vital, not >>>unnecessary knowledge, so that the released capacity has driven the search >>>another ply, or at least part of a ply. This applies to Shredder too, of course. >>> I remember some time ago the author of Hiarcs saying that he had vastly >>>increased the knowledge in his new version. This could only have come at the >>>expense of search depth, and so was a mistake. >>> Just one other point re: Fruit2.1. I could'nt help notice that Fruit2.1 has a >>>preference for knights over bishops. This I noticed over a series of too many >>>games played on Arena for it to be random. I'm only guessing, but could it be >>>that the knight, handled correctly, is a stronger piece overall than a bishop? >>>And that this has not been recognized in the chess community until now simply >>>because it has been impossible to search deep enough to demonstrate? Food for >>>thought. >>> all the best >>> Lin >> >> >>Hi Lin, >> >>nice to hear from you, as always. Yes, Fruit might be the currently optimal way >>to conduct search, perhaps even more so than any other program. > >I disagree > >Fabien already said that fruit has search bugs and the main improvement from old >fruit to newer version is the evaluation. > >I guess that Fabien may do Fruit 100 elo better by only improving the search by >improving things like order of moves and avoiding useless extensions. > > > I guess that >>Fabien will have to modify it a little to make it even stronger. Alas, there >>may be a rub there, as any addition to the optimal search is dangerous and may >>upset the overall strength. > >Fruit's search is not optimal. >It may be better than the poor search of most programs but it is not close to be >optimal. The optimal search has a branching factor of 1.000 I don't think we will see it any time soon. >>Chess System Tal comes to my mind as the knowledge-laden program that had >>promise. > >I doubt if Chess system tal has better knowledge than fruit. > > > However, it appears now that only increased depth and better search >>point the way to go. CST would be decimated in its match with Fruit, as, >>unfortunately, is the case with any knowledgeable program, no matter how >>attractive its playing style may be. >> >>Simulating human chess knowledge is not the way chess programming should go, it >>slows down the search so much so that such programs overlook some important >>tactics. > >I agree that simulating human chess knowledge is not the way to go for the >simple reason that humans do not know a lot of things including not knowing how >they think. It may or may not be a good solution. I don't think we really know the answer to that question one way or the other. >I disagree that more knowledge is not the way to go but it is important to check >that the knowledge that it implemented is productive and help the program to >play better. >> >>I have noticed Fruit's preference of knights over bishops too, as well as its >>occasional tendency to sac a light piece for a couple of passer pawns. I have >>also noticed that it undervalues rooks... These could be interesting pointers >>for Fabien... >> >>Thanks. > >I did not notice that Fruit prefers knights over bishop and I found that in one >position from my correspondence game it is Fruit that likes to keep pair of >bishops and not Shredder. An important thing to remember is that search is clearly a form of knowledge. This has been formulated in clear terms by Christophe Theron. It is trivially obvious that if you search clear to win/loss/draw then your engine has perfect knowledge. Search and purely static understanding of the position both arrive at the same thing: Better information for choosing the move.
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.