Author: Vincent Diepeveen
Date: 19:01:55 08/27/02
Go up one level in this thread
On August 27, 2002 at 18:40:55, Uri Blass wrote: 2 ply what a joke. I get usually in champs outsearched by 3-4 ply by my opponents. >On August 27, 2002 at 17:57:50, Uri Blass wrote: > >>On August 27, 2002 at 17:48:35, Vincent Diepeveen wrote: >> >>>On August 27, 2002 at 07:57:31, Uri Blass wrote: >>> >>>>On August 27, 2002 at 07:36:44, Arturo Ochoa wrote: >>>> >>>>>On August 27, 2002 at 07:12:04, Uri Blass wrote: >>>>> >>>>>>On August 26, 2002 at 17:37:56, Arturo Ochoa wrote: >>>>>> >>>>>>>On August 26, 2002 at 14:44:47, Uri Blass wrote: >>>>>>> >>>>>>>>On August 26, 2002 at 13:53:14, Dann Corbit wrote: >>>>>>>> >>>>>>>>>http://f11.parsimony.net/forum16635/messages/33526.htm >>>>>>>> >>>>>>>>It is better to play with the program against other opponents and not against >>>>>>>>itself to get a good estimate and not to play more than 2 games in a >>>>>>>>match(otherwise the problem can be aggresive learning and not lack of book). >>>>>>> >>>>>>>Hello: >>>>>>> >>>>>>>No, it is not better and it doesn't have any sense. As you declared in the long >>>>>>>thread below, you believe (but it is not demonstrated) that a engine with book >>>>>>>is not better than a engine without it. >>>>>> >>>>>>I did not say that engine with book is not better than an engine without it. >>>>>> >>>>> >>>>>>For you it is impressive to see the same game again and again. >>>>>>Not for me. >>>>> >>>>>The thing is to prove that a Book helps an Engine a lot to improve its level >>>>>during the Opening. >>>>> >>>>>The idea is not to have random books because it is not interesting. >>>>> >>>>>I agree that to say impressive must no be mentioned here. >>>>> >>>>>But, you seems to misvalue the remarkable fact of a tuned book for a chess >>>>>engine. >>>>> >>>>> >>>>>> >>>>>> >>>>>>chess programs cannot do it but they have the potential to do it if programmers >>>>>>improve them so I think that it is better to look at the positions when programs >>>>>>blunder in the opening and to make the right observation how to improve programs >>>>>>based on looking in these positions. >>>>>> >>>>>>Movei with almost no knowledge can find book moves by itself in big majority of >>>>>>the cases so I can only imagine what a chess engine with clearly better >>>>>>knowledge about search rules and about evaluation can do. >>>>> >>>>>Yes, then I should conclude that every engine including Pierre, MSCP, etc. can >>>>>solve the opening problems without almost any knowledge. >>>>> >>>>>If this were real true, I should ignore 50 years of theory. >>>>> >>>>>I don´t know what thing is called "Movei". I suppose is your private program or >>>>>so. If you declare that this program without knowledge can solve the big >>>>>majority of the cases, I would like to know what is "Big majority": 90%, 95%, >>>>>99%. >>>> >>>>I guess 90% >>>> >>>>> >>>>>I would be admired to believe that this program is competing at the same level >>>>>of Gandalf, Yace or Insomniac without requiring a tuned book? >>>> >>>>No >>>>It is not close to the level of yace >>>> >>>>A previous version that is at similiar level to the last version with a very >>>>small books of less than 1 kbytes(I have one book for white and one book for >>>>black) >>>>lost 40-10 against yace that used only defensive learning to avoid losing the >>>>same game twice. >>>> >>>> >>>>> >>>>>Are realizing about you are saying? The programmer should see the position where >>>>>the engine is blundering? >>>>> >>>>>This cannot solve the problem because you can have 1000000 positions where the >>>>>programs, "the Top Programs" and the amateurs blunders a lot and it is not a >>>>>problem of solving the position only. It is the problem of all the problem: >>>>>strategy. >>>>> >>>>>Tell me: how a program without knwoledge can solve strategical problems of the >>>>>openings. >>>>> >>>>>I would be admired. >>>> >>>> >>>>Strategy is often tactics+right evaluation if you search deep enough. >>> >>>In fact by definition it is the case that if evaluation is right >>>you only need 1 ply search depth to find everything :) >>> >>>However we all know, except Uri, that programs suck ass strategical and >>>positional. Of course Uri, >>>not playing himself at any significant level himself, and probably >>>himself always losing because he drops a piece somewhere; if you can't >>>understand that there are many games in this world which do NOT get >>>decided by dropping a piece, then obviously debating further makes no sense. >> >>Of course I understand that there are games that are decided not by material >>mistakes and I did not say that only material evaluation is enough >>or even that only piece square table is enough but the point is that >>often it is possible to translate positional advantage that >>the program does not understand to positional understanding that the >>program does understand. >> >>If you think that movei always like material then you have a big >>mistake. >> >>After 1.e4 d5 exd5 Nf6 movei plays by search d4 and not c4 inspite >>of the fact that c4 earns a pawn. > >In the opening and the middle game in some cases. > >Movei has a very small evaluation today(less than 200 lines of C) >but it does not prevent it to beat some amateurs with bigger >evaluation thanks to better positional understanding. > >The evaluation of the top programs is better in most of the cases >and they have also >better search rules and better use of hash tables but I believe that >if you give movei >advantage of 2 effective plies relative to the top programs >then it can compete at least as equal against them in the >middle game(even at 120/40 time control for the top programs). > >The problem is that movei may need some hours for every move >in order to get these effective 2 plies advantage. > >It need more time than 3 minute per move only in order to be equal >in effective search depth. > >effective search depth is not the number of plies. > >It is possible to get an estimate for it by taking positions when >programs won material from games and needed more than a second to >fail high and calculate average depth >for movei to fail high and for the top programs to fail high. > >If the average depth of Hiarcs is 10.3 and the average depth of >Movei is 12.3 then movei needs to search 2 plies deeper than >hiarcs to get the same effective depth and 4 plies deeper than hiarcs >to get the advantage that I talk about. > >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.