Author: Vasik Rajlich
Date: 04:58:24 09/26/04
Go up one level in this thread
On September 25, 2004 at 15:16:52, Sandro Necchi wrote: >On September 25, 2004 at 14:52:42, Vasik Rajlich wrote: > >>On September 25, 2004 at 11:19:52, Sandro Necchi wrote: >> >>>On September 25, 2004 at 11:12:19, Vasik Rajlich wrote: >>> >>>>On September 25, 2004 at 09:47:23, Sandro Necchi wrote: >>>> >>>>>On September 25, 2004 at 03:57:30, Uri Blass wrote: >>>>> >>>>>>On September 25, 2004 at 01:56:37, Sandro Necchi wrote: >>>>>> >>>>>>>On September 24, 2004 at 13:05:52, Sandro Necchi wrote: >>>>>>> >>>>>>>>On September 24, 2004 at 12:09:00, Vasik Rajlich wrote: >>>>>>>> >>>>>>>>>On September 23, 2004 at 13:31:55, Sandro Necchi wrote: >>>>>>>>> >>>>>>>>>>On September 23, 2004 at 01:44:08, Uri Blass wrote: >>>>>>>>>> >>>>>>>>>>>On September 23, 2004 at 01:31:37, Sandro Necchi wrote: >>>>>>>>>>> >>>>>>>>>>>>On September 22, 2004 at 06:58:33, martin fierz wrote: >>>>>>>>>>>> >>>>>>>>>>>>>On September 22, 2004 at 05:56:02, Vikrant Malvankar wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>>It is not a benefit for a weak engine as it will also probably play weak moves >>>>>>>>>>>>>>in the middlegame which will be properly exploited by the stronger engine. Dont >>>>>>>>>>>>>>u think so. >>>>>>>>>>>>> >>>>>>>>>>>>>it's not the issue whether a strong engine will beat a weak engine. that is so >>>>>>>>>>>>>by definition :-) >>>>>>>>>>>>> >>>>>>>>>>>>>the question is: take 2 engines of approximately equal playing strength, give >>>>>>>>>>>>>one of them a good book, and look what happens in a match. >>>>>>>>>>>>> >>>>>>>>>>>>>i believe that for 2 weak engines the difference will be larger in the match >>>>>>>>>>>>>result than for 2 strong engines. >>>>>>>>>>>>> >>>>>>>>>>>>>now we only need somebody to test this hypothesis :-) >>>>>>>>>>>>> >>>>>>>>>>>>>cheers >>>>>>>>>>>>> martin >>>>>>>>>>>> >>>>>>>>>>>>Hi, >>>>>>>>>>>> >>>>>>>>>>>>I made very many tests and I can make statements on this matter: >>>>>>>>>>>> >>>>>>>>>>>>1. A program stronger 150 points than another will win nearly all games no >>>>>>>>>>>>matter how bad it comes out from the openings. >>>>>>>>>>>>2. The stronger the program is the most important the book is. Of course weak >>>>>>>>>>>>lines should be checked and removed to avoid loosing positions. >>>>>>>>>>>>3. The weaker the program is the less the book is important. The reason is that >>>>>>>>>>>>it will find very many positions where it does not know how to play them. >>>>>>>>>>>> >>>>>>>>>>>>P.N. Do not take the Shredder - Hydra example to state the opposite, because I >>>>>>>>>>>>knew we had some weak lines in the book, but for personal reasons could not work >>>>>>>>>>>>on them. >>>>>>>>>>>> >>>>>>>>>>>>Of course anybody can state the opposite, but my statements are supported by >>>>>>>>>>>>thousand of games and more than 100 engines/prototype testing at all level and >>>>>>>>>>>>with very many different harware. >>>>>>>>>>>> >>>>>>>>>>>>I have no time and williness to do deeper into these matters, so it is up to you >>>>>>>>>>>>to believe me or not. >>>>>>>>>>>> >>>>>>>>>>>>Sandro >>>>>>>>>>> >>>>>>>>>>>At the very weak level books are not important because the program that get >>>>>>>>>>>better position cannot use it. >>>>>>>>>>> >>>>>>>>>>>At the very high level books are also not important because the program can find >>>>>>>>>>>better moves by itself. >>>>>>>>>> >>>>>>>>>>No, this is today totally wrong in at least 95% cases. >>>>>>>>>> >>>>>>>>>>It depends on the positions, but in some positions they should search at 64/108 >>>>>>>>>>to be able to do it and I do not think any chess program is able to reach those >>>>>>>>>>depths now. >>>>>>>>>> >>>>>>>>>>I have made several tests running fast harware for more than one day and the >>>>>>>>>>moves and the evaluation they got was poor compared to real ones. >>>>>>>>> >>>>>>>>>Depends on what "real ones" means. Humans also make mistakes. >>>>>>>> >>>>>>>>Yes, but I was referring to deep analysis of a position, not games. Some times >>>>>>>>deep analysis takes days, months or even longer...otherwise is not deep...:-) >>>>>>> >>>>>>>An example: >>>>>>> >>>>>>>after 1. e4 c5 2. Nf3 d6 3. d4 cd4 4. Nd4 Nf6 5. Nc3 a6 can computers answer >>>>>>>these questions: >>>>>>> >>>>>>>1. Is this the best line for white? >>>>>> >>>>>>I guess that humans cannot answer better. >>>>> >>>>>They do...ECO volums are about 80% reliable. Even if there is still a lot of >>>>>room for improvements this does not mean that the computers will improve theory. >>>>> >>>>>Just consider what follows: >>>>> >>>>>1. Current theory is based on more than 100 years games played by many strong >>>>>players also at corrispondance chess too. >>>>>2. Top engines do not have more chess knowledge of GMs and they do not see >>>>>deeper. A program looking at 18/44 see 9 full moves as average, 22 on the best >>>>>line and maybe 4-5 on some lines which are cut early. This is not much compared >>>>>to deep analysis by GMs. >>>>> >>>>>If the programs cannot use the knowledge made by strong players they need to see >>>>>quite deeper that they can. Of course there are positions where they look ahead >>>>>is already enough and on these they are dangerous players, but these are mostly >>>>>tactical ones or where material gain is an important factor. >>>>> >>>>>Positions where a positional compensation for given material is important are >>>>>not handled well unless the look ahead can see how to get back material and or a >>>>>mate. >>>>> >>>>>Of course weak reply can make weak moves very strong, but it depends who is the >>>>>opponent... >>>>> >>>>>> >>>>>>>2. Is 2...d6 best move for black? >>>>>> >>>>>>Again I guess that humans cannot answer. >>>>> >>>>>They gave a 80% reliable answer...the future will see this percentage raise more >>>>>and more... >>>>> >>>>>>Probably 2...d6 is one of some drawing moves but I cannot be sure about it. >>>>>> >>>>>>>3. Is this line best line for black? >>>>>>>4. What is white best move at move 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, >>>>>>>18, 19 and 20? >>>>>>>5. What are the best reply for black on those moves and the white best line? >>>>>>>6. How deep should a chess program need to search to give these answers? >>>>>>> >>>>>>>Uri, do you really think a chess program can give better answers (moves) than a >>>>>>>strong human player? >>>>>> >>>>>>I do not know. >>>>>>I think that in most cases they will give moves with the same quality. >>>>>>In some cases espacially in moves 11-20 they may give better moves if you give >>>>>>them a long time to analyze. >>>>> >>>>>My very many tests show that this is happening very seldom. >>>> >>>>The big question is if the engine's moves are really worse, or just different. >>> >>>Most of the time worse...a few times different. >>> >>>Most of the time worse because the horizonth effect is giving not engough >>>depth...a few times different because they do not follows rules as humans do and >>>look also moves which will be normally weak, so humans do not investigate on >>>them usually... >>> >>>Sandro >> >>In this case, a competent correspondence chess player should crush an engine >>running without human assistance. > >I agree. > >> >>IMO a lot of people don't trust computer moves, because they often look so ugly >>- but in fact those moves are actually not so weak. > >Of course not so weak, but a GM would see that they are not the best and >increase the advantage move after move...normally a GM makes at least 2 mistakes >and some minor ones in a game, but on correspondance chess it would be >different... > Yes - but sometimes the computer's move is in fact better, and the humans are wrong to override it. Consider the following position: [Event "BGN World Chess Championship"] [Date "2000.10.10"] [Round "2"] [White "Kramnik, Vladimir"] [Black "Kasparov, Garry"] 1. d4 Nf6 2. c4 g6 3. Nc3 d5 4. cxd5 Nxd5 5. e4 Nxc3 6. bxc3 Bg7 7. Nf3 c5 8. Be3 Qa5 9. Qd2 Bg4 10. Rb1 a6 [D] rn2k2r/1p2ppbp/p5p1/q1p5/3PP1b1/2P1BN2/P2Q1PPP/1R2KB1R w Kkq - 0 11 Here Kramnik sprung a very strong novelty - the previously ignored "computer move", 11. Rxb7. Kasparov was not ready for it, and did not repeat the Gruenfeld in the rest of the match. Vas >Sandro >> >>Vas >> >>>> >>>>Vas >>>> >>>>>> >>>>>>Uri >>>>> >>>>>Sandro
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.