Author: Vincent Diepeveen
Date: 14:50:06 01/17/99
Go up one level in this thread
On January 16, 1999 at 15:36:04, blass uri wrote: > >On January 16, 1999 at 15:14:25, blass uri wrote: > >> >>On January 16, 1999 at 15:03:33, Vincent Diepeveen wrote: >> >>>On January 15, 1999 at 08:48:28, Robert Hyatt wrote: >>> >>>>On January 15, 1999 at 02:26:01, Kai Skibbe wrote: >>>> >>>>>On January 14, 1999 at 22:55:15, Robert Hyatt wrote: >>>>> >>>>>>I'm willing to do most anything, but I have no idea what is needed, since >>>>>>I don't know how they read the book, how their format looks, etc. IE it >>>>>>seems odd to have a book algorithm that works perfectly well already built >>>>>>into crafty and it can't use it... >>>>> >>>>>Hello Bob, >>>>> >>>>>why don´t you patch your winboard-engine to work with fritz ? The engine is >>>>>already running with full tablebase-support and using the crafty-book. The major >>>>>problem still is, that at every move the hashtables are cleared. But I think >>>>>this problem could be solved easily. Roland Pfister, the author of Patzer, has >>>>>already patched his code of his winboard-engine, so that the hashtables are not >>>>>cleared. I think patzer reocognizes the "fritz"-command, which is send to the >>>>>engine at the startup. >>>>> >>>>> >>>>>Best regards >>>>>Kai >>>> >>>>Firstly, I'm not a 'windows' programmer and don't know what has to be done to >>>>work with the fritz API. Hash table clearing is only one problem that I have >>>>described, but there are others. IE I carry lots of things from one search to >>>>the next, from killer moves, to history counts, to the PV from the last search, >>>>to the pawn hash stuff, and all of this gets wiped out when a bunch of moves >>>>are 'stuffed' into crafty. >>>> >>>>I at least assume that the interface makes sure that both engines are not >>>>'pondering' or using 'permanent brain' (same thing of course) as it is very >>>>possible to turn it off in one, and leave it on in the other, and get some >>>>nasty results... IE if crafty 'ponders' it does no good since the move list >>>>gets stuffed in when the opponent moves. But it would certainly affect the >>>>match because it would eat 50% of the cpu while the opponent was thinking, >>>>and then throw that result away. So it would turn a PII/400 into a PII/200 >>>>as far as the other program was concerned. And it would make a difference in >>>>how things turned out. >>>> >>>>This is a mess, to say the least... >>> >>>Not really a mess. Chessbase spends time, so chessbase gets back result. >>>That's how i see it. >>> >>>Their result is that other engines slowdown and suffer from the limited >>>command set, with the result that they are twice as fast at the same >>>computer when autoplaying there. That is a clear result. As matthias >>>Feist is a very good programmer (just look how well interface actually is) >>>we can only ASSUME it is done not by accident, but because it was demanded >>>by someone. That someone is of course the one paying him. The one paying >>>him is chessbase. >>> >>>It is simply a way to make fritz performance look better, that's how i >>>see it. Of course what would you do? Crafty is freeware, fritz is >>>a commercial program, so it is in their interest to do things in a >>>way which by some can be seen as unfair. It's very doubtful whether it is >>>unfair. If i tell this to a programmer then he says: unfair. >>> >>I do not think they earn by unfair things >>I do not buy fritz5.32 before they do the real crafty(with tablebases and >>without clearing the hash tables) an engine for Fritz5.32 >> >>Uri > >I also believe that the engines do not use permanent brain when they play one >against the other. At my PII-450 dual they all use permanent brain when i do 2 machine mode. Not sure what they do at your place. >If they were using permanent brain then I could see in games between engines for >Junior5 that one engine is at big depth after a few seconds and I did not see >it. progs that rely on piece square table like Rebel, Fritz5, Junior all need to clear hashtables before starting a search. There is a clear reason for that. if you adjust values then hashtable messes them up as the previous time you searched the an evaluation of a position was different. Crafty, DIEP and most other programs i know however use evaluation in the leaf. Bob can explain this more clearly to you than i can. Greetings, Vincent >I can also say the engines not to use permanent brain in the engines parameters > >I assume that the bugs were done by accident also because I found other bugs >that did not help them in their programs. >I wait to hear if all the problems are solved. >I understood that the problem of clearing the hash tables is solved. > >Uri
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.