Author: Ralf Elvsén
Date: 14:59:54 07/04/00
Go up one level in this thread
On July 04, 2000 at 17:14:47, blass uri wrote: >On July 04, 2000 at 16:35:06, Ralf Elvsén wrote: > >>On July 04, 2000 at 08:04:25, blass uri wrote: >> >>>On July 04, 2000 at 06:56:07, Ralf Elvsén wrote: >>> >>>>On July 04, 2000 at 04:28:36, blass uri wrote: >>>> >>>>>On July 04, 2000 at 02:34:55, Steve wrote: >>>>> >>>>>>On July 03, 2000 at 18:49:14, Torstein Hall wrote: >>>>>> >>>>>>>I'm not so sure that Junior will give better analysis than Fritz 5. They are >>>>>>>both fast searcher with about the same level of pos. knowledge. Anyway I feel >>>>>>>that all the programs give most tactical variations anyway, even Hiarcs 7.32, >>>>>>>considered among many as the most positional engine. >>>>>> >>>>>>I agree. Hiarcs 7.32 is great, but its strength (just like every other >>>>>>engine's) is tactics. Its analysis of endgames is often horrendous. >>>>>> >>>>>>> >>>>>>>I rather go for Crafty! Its free, has more chess knowledge than most of the >>>>>>>programs and runs as a native chessbase engine! >>>>>>> >>>>>>>Torstein >>>>>> >>>>>>A number of people have made the same observation in this forum. But whenever I >>>>>>play Hiarcs 7.32 against Crafty (running as ChessBase engine) -- generally at >>>>>>either G/30 or G/60 -- Hiarcs wins easily. >>>>> >>>>>The only game that I played Hiarcs7.32 against Crafty17.11 at 3 hours/40 moves >>>>>Crafty won and my observation was that hiarcs has a bug at time control that is >>>>>longer than G/30 or G/60 and this is probably the main reason that it failed in >>>>>the ssdf list. >>>>> >>>>>The bug is in playing and hiarcs could probably be a better program in >>>>>tournament time control if it knew to clear its hash tables after every move >>>>>because learning from previous search is bad if it is done in the wrong way. >>>>> >>>>>Uri >>>> >>>>I think you are too quick to draw conclusions. This "strange-moves-from- >>>>Hiarcs-that-can't-be-reproduced-by-others-bug" is well known, but how >>>>frequent is it? I have observed plenty of long time control games by >>>>Hiarcs 7.32 and it is very seldom I suspect this to happen. But who knows? >>>>I guess one would have to look carefully at every game afterwards. My >>>>personal quess is that keeping the hash tables is better, but I have >>>>no "proofs" . >>>> >>>>As to the strength of Hiarcs and Crafty, I have my own firm opinion >>>>but it is "unscientific" and I'll keep it to myself :) >>>> >>>>Ralf >>> >>>The problem is mainly long time control problem. >>> >>>Hiarcs won Crafty17.11 3.5:.5 in my 40/40 games on one pentiumIII450 but this >>>time control is faster than tournament time also because of the fact thatthe >>>sides cannot ponder when 3 hours/40 moves on one computer is similiar to 2 >>>hours/40 moves on 2 computers. >>> >>>There are cases when the problem does not lead to strange moves but lead to a >>>longer "think" by hiarcs and it also can cause later mistakes on time trouble. >>> >>>In the case of the game that I played between hiarcs and crafty hiarcs did not >>>blunder by Bxb6 but needed a long time to avoid it(at longer time control it >>>even plays Bxb6). >> >>Yes, I didn't read your post carefully enough. Anyway, I wonder how often >>this happens. This, the other bug and some serious misevaluations Hiarcs >>does is something I hope will be worked out in the next version. >>I like the program alot but have always felt it could be significantly >>stronger if tuned better. If Mark Uniacke posted here I could produce >>a pretty long list with "complaints". Eh, no wonder he doesn't... >> >>It's not just the wellknown king safety problem. E.g. the value >>of pieces must be modyfied according to the position. It has lost >>many games because of this. >> >>Ralf > >I think that another problem is the limit for the selective depth of hiarcs. >Hiarcs can never see more than 31 plies forward and it may be also a problem in >long time control. > >I think that you will have a long list of complains to every programmer because >finding problem is easier than fixing them and fixing one problem may cause >another problem. Hmm, yes, I have thought about this. But there isn't much discussions about Shredder or Tiger although their authors post here. Maybe these programs are perfect :) I am reluctant to give examples where Crafty misevaluates positions since there are plenty of Crafty posts around anyway. > >I believe that the problems that I posted about some wrong learning from >previous search and in this post about the maximal search depth are more easy to >fix without creating other problems. > >Uri I think I can guesstimate well enough what is doable and not since I'm trying to program myself, with very moderate success, and have studied some of Crafty's source code. Take Hiarcs evaluation of rooks vs two light pieces, or two rooks vs queen, depending on the position. I just can't see why this would be so hard to fix. But I agree, it's easy to complain, although I should have used the expression "try to help" . Oh well... Ralf
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.