Author: Thorsten Czub
Date: 17:23:53 11/08/00
Go up one level in this thread
On November 08, 2000 at 10:06:15, Uri Blass wrote: >I think that you cannot find it by big evaluations because big evaluations can >help programs to lose by wrong sacrifices. ?!? not if the knowledge that produces the big results is good enough to differenciate between "right" or "wrong". >It happened to Gandalf4.32 that lost against shredder2 on slower hardware >because of speculative sacrifices. aha. let us see... >Here is the relevant position again: [D]3rr1k1/1bq1b1p1/pp1ppnnp/2p5/2P1P3/1PNBBN1P/P2Q1PP1/R2R2K1 w - - 0 1 >Gandalf sacrificed by Bxh6 and lost the game(I am not sure if Bxh6 is wrong but >gandalf played weak after this move). CSTal2.03 wants to play Bxh6 with a score of +1.74. after a while d10, 121s, +1.93, Bxh6 Nh5 Bg5 Rf8 Bxe7 Qxe7 Na4 as usual :-)))) cstal does not want to take the bishop ! :-)))) >The line of the game was 20.Bxh6 gxh6 21.Qxh6 Nf8 22.Ng5? after gxh6 CSTal2.03 says d9, 28s, +2.71, Qxh6 Nf8 e5 Nh7 Bxh7+ Nxh7 d10, 82s, +2.61, Qxh6 Nf8 Ng5 e5 Nf3 Rc8 Nh4 ... after Nf8 i get d9, 84s, +2.74, Ng5 e5 Nd5 Nxd5 Nxe6 Nxe6 Qxe6+ Kh7 >Gambittiger shows the same main line but it can after more than 12 minutes on my >pentiumIII450(192Mbytes hash) to avoid Bxh6 when Gandalf cannot like Bxh6 even >after 14 hours. IMO you make a mistake. you think there are programs and programs. and the one programs don't sac. and the others sac. and those saccing have to be optimized. the way that their sacs work. but - the mistake is here: a working sac, IS NO SAC at all. a working "accurate" sac means: the opponent has made a mistake BEFORE, so the position is WON and only the person/side to move has to find the key move (spectacular = sac!!). but new paradigm is not about finding key moves in positions that ARE WON. thats what i call crossword-puzzle-solving. River with 3 letters, very long. the solution is there. why call it sac ?? what do you sac ? there is no risk. no fuzzyness. its clear the position is won. so if your move makes the side to move win the game, there is no sacryfice. You believe chess is a kind of Bednorz-Toennissen test-suite, and the program saccing only has to find the "right" move to win. i don't think this is a sac per definition. this is NOT a sac. there is often misunderstanding about those terms. e.g. many people call their relationships to others love, when it fact it is symbiotic ! you give something, you get something. thats not love. thats symbiotic. thats exchange of goods. but not love. when you love somebody, you give something, and you don't expect to get anything from this person, but you do it because it makes YOU feel good. it makes you happy to give. but is does not really have any logical or material reason. you somehow SAC something, its a kind of investion. with a risk. you trust that it will work. but nobody knows. you can fail. and lose. and lose your money you invested, or whatever you gave. but you have to do it. if you define sac= i give something and therefore get something, this is not a sac, i would call this a combination. a tactical or (if long range) a strategical one, but not a sac. >I will test later if it can avoid Ng5 as you see cstal would play the same "sac". now you research if the move is correct ? why ? i don't have shredder2 anymore. only shredder4. ok - shredder4 wants to play Bxh6 too. so maybe ... the problem here is not a problem of beeing speculative or not. maybe the plot of the position is too deep for most of the programs. after a while it sees a draw. than again it believes white is better. >I think that big evaluations can be productive but you should do tests to see >that it is not big in the wrong places. this is IMO the wrong idea. you believe that there are exceptions. i remember ossi weiner saying: oh richard. don't wory. the rule does not work always. it's an exception in this position. here the rule (never exchange your fianchetto-bishop when the opponent has the bishop of the same color AND the queen on board) is not working. >A good test suite about evaluation of program should have cases when speculative >sacrifices are wrong and not only cases when speculative sacrifices are right. this is true. but IMO the problem is not that the programs always sac. or do NOT sac. do they PLAY and PLAN for the sac. if you do not only wait until a position is there (because the opponent has too often overseen something and the position is fruitful) but actively do something to make the position interesting. because IF you do so, you normally cannot go back. thats the interesting thing. cstal tried to make an attack step by step. it really invented them. IMO gambit tiger is doing the same. same for gandalf. both do it not always. but often. you come with a position that is unclear. and you say: oh - here it is a danger. it does not work against shredder2. i do not believe in these ideas. imo the programs should not try to find out HOW a position is to evaluate right. they should IMO better drive directly into the area they understand the most. this is what i do in life. when i understand something about AMD, i do not build the customer an Intel pentium machine :-)) when somebody wants to talk with me about a product, i try to control the talk. because i can than direct the customer to the parts i have and not to parts i do not have. if i lose this control, i will not sell something. the same with chess programs. if the program is unable to control the game (with fritz or other bean-counters: they cannot control because they don't know about chess/~products) it can happen that it loses or the opponent makes a draw out of your won position. this was the problem with tiger12. it played nice. but... it was unable to control the game. kramnik was able to control the game in many games against kasparov. only he had problems to win the winning positions. to control the game, you need to know (which products you have in stocks/ you know about) WHAT you know. i cannot allow to let the customer talk about a graphics card i don't have and don't know about the specifications or the incompatibility. i have to FORCE him to buy the graphics card I have, i know it is good, and i know it will not make incompatible problems with HIS hardware. that is what cstal and gambit-tiger do. it is not important if the moves are accurate or correct. nobody knows. you have to believe ! if you elect bush or gore, you don't know what will happen, if the guy has told you really what he will do, or wants to do. or if it will be real. you have to believe in it. trust ! you don't get something back. you have to sac in forward. you cannot proof if your decision was right. only later, much later, after the game, after years, ... can you follow me until here ?! ><snipped> >>right. >>bob is on the wrong track. >>he searchers for help. but i think there is no help in that fight anymore. >>gambit tiger is there. you have to live with it. >>you cannot ignore gambit tiger anymore. >>and not the way it gets its results. >>thats part of it. >1)Gambittiger wins in part of the games because it is good in the endgame. >Gambittiger is also better than Crafty in tactics. right. >The fact that Gambittiger wins against Crafty is not a proof that it is because >of better king safety evaluation because when you are better in tactics playing >for king attack can help you to score better. right. the difference is that gambit tiger plays for the things it understands. like any human does. it's stupid NOT to play/plan for the things you are capable of. if you are good in playing a piano, and you play cello, and fail, than your decision to play piano was shit. >The real test is against tiger13 that is not weaker than Gambit in tactics. >I guess that gambit is going to beat also tiger13 in the future because >christophe only started the work about gambittiger and he is going to improve >it. right. but you are mistaken by believing that you can measure the strength of an engine by let it play against a version before or after. this is NOT an absolute method to make progress. 2 programs that have nearly the same basement, will fight against each other, but the result will not tell you much about their real playing strength. only by doing fights against all opponent programs test-suites test games against humans you can be sure. >2)Crafty lost in your games against gmabittiger but Crafty could not use more >than one processor. true. i don't have a multi-cpu-machine. >My guess is that Bob is going to care about tiger when Crafty is going to lose >against Gambit on ICC. :-)))) that sounds very logical to me. yes. i forgot that bob is only a human beeing too. :-)) >I guess that he cares more about Deep Fritz because Deep Fritz scores better >against crafty on ICC. brilliant. i guess this is the reason because computerchess makes no progress. incest. and the way we handle it... >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.