Author: Uri Blass
Date: 18:10:20 05/07/02
Go up one level in this thread
On May 07, 2002 at 20:24:40, stuart taylor wrote: >On May 07, 2002 at 13:20:28, Uri Blass wrote: > >>On May 07, 2002 at 12:53:13, Vincent Diepeveen wrote: >> >>>On May 06, 2002 at 13:09:45, Christophe Theron wrote: >>> >>>>On May 06, 2002 at 01:05:23, Uri Blass wrote: >>>> >>>>>On May 05, 2002 at 19:58:09, Christophe Theron wrote: >>>>> >>>>>>On May 05, 2002 at 19:25:07, stuart taylor wrote: >>>>>> >>>>>>>I mean, where are we? I cannot make it out yet. >>>>>>>Can we safely say that a top program of today can beat all programs from before >>>>>>>1996, i.e. 1995 and below? >>>>>> >>>>>> >>>>>>Year of release? >>>>>> >>>>>>Wrong thinking. >>>>>> >>>>>>What you need to consider is the number of years the programmer has spent >>>>>>actively working on his program. >>>>>> >>>>>>That gives much better figures. >>>>>> >>>>>>Genius 5 is a program of 1996, but it represents approximately 15 years of hard >>>>>>work by Richard Lang. >>>>>> >>>>>>Now consider an amateur program of 2002, on which the programmer works since >>>>>>1996. >>>>>> >>>>>>Are you going to compare 1996 and 2002 and decide that the 2002 program is >>>>>>probably better? >>>>> >>>>>The amateur of 2002 has the advantage that the programmer could get more ideas >>>>>about programming from reading and also could do better testing thanks to better >>>>>hardware and software. >>>>> >>>>>I can give you one example for the last point and it was about testing to find >>>>>bugs in my move generator: >>>>> >>>>>There are a lot of programs that calculate today the perft function for every >>>>>position(perft 6 is the number of legal games of 6 moves from the position) . >>>>>They helped to find bugs in my complicated move generator(if I see that perft 5 >>>>>is not correct then I can find the bug by finding a position when perft 4 is >>>>>wrong,finding a position when perft 3 is wrong...). >>>>> >>>>>I guess that many years ago there was no free software to calculate that >>>>>function and even if there was software to do it the hardware caused it to be >>>>>clearly slower so testing and finding bugs was an harder task. >>>>> >>>>>>about when they talk about knowledge) makes for 10% of the strength of a chess >>>>>>program. >>>>>> >>>>>>Chess is 90% about tactics (which is a concept close to "search"). >>>>> >>>>>It is possible that evaluation may be more important but programmers failed to >>>>>write the right evaluation to prove it. >>>> >>>> >>>>You are right. >>>> >>>>I consider that current top chess programs, compared to top human chess players, >>>>are at the 50% level in search and 20% in evaluation. >>>> >>>>I do not even know if it is 50/20, considering that top human ches players, who >>>>are thousands of times slower, can equal or dominate the top chess programs >>>>running on top hardware. >>>> >>>>So there is ample room for improving, that's the good news. >>>> >>>>But I still believe that the proportion in chess is 90% importance for search >>>>and 10% for evaluation. >>>> >>>>There is more room for improvement in evaluation (in percentage relative to the >>>>skills of human players), but it still represents a minor part compared to >>>>potential improvements in search. >>>> >>>> >>>> >>>> Christophe >>> >>>for chessprograms things are different. Not a single one of them >>>misses obvious tricks nowadays. It's book + evaluation dominated simply. >> >>The main advantage of searching deeper is not obvious tricks. >>It can be better positional moves and it can be tricks that are not obvious. >> >>I think that in most comp-comp games between equal programs if you give the >>losers to search 2 plies deeper when the winner search to the same depth that >>they searched in the game then you are going to see different results. >> >>book dominates? >>I think that most programmers are going to prefer to play with no book if their >>hardware is 3 times faster. >> >>I think that it includes also christophe but christophe can tell if I am wrong. >> >>Uri > >Uri, this is a big Chidush! Do you mean 3 times as fast, or 3 plies deeper? I mean 3 times as fast. >There is quite alot to know in order to made opening decisions without a book. I >don't see how x3 hardware speed will be enough for that >S.Taylor My program(movei) plays in the 5th division with no book and it gets good results there(20.5/24 inspite of having no books when a lot of opponents have book). The result could be even better if it was better in tactics(for example it blundered and lost a pawn in the opening against the best program of the 5th division resp and null move that was not used in the 5th division let it to search one ply deeper). You can download the games in the following link when you choose the 5th division http://home.hccnet.nl/leo.dijksman/index.html I can also show you that Deep Shredder played in a tournament with no book and got good results(it did not got first place and I did not say that book means nothing but it is clearly better than most of the amatuers with no book). see http://home1.t-online.de/home/g.simon.rgbg/news.htm I think that a program can often make the right choice in book positions by itself and even if it does not do it then it can often defend the slightly worse position by itself in order to get at least a draw. There are possible tactical traps that a program with no book can fall into them but if the opponents do not know the evaluation of the program without book they cannot plan their win before the game and if the programmer can change the evaluation between games then he can avoid falling in the same trap twice. It is easy by a small change in the evaluation that almost does not change the playing strength to change the first move from 1.e4 to 1.d4. 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.