Author: Uri Blass
Date: 11:43:41 12/30/03
Go up one level in this thread
On December 30, 2003 at 14:22:24, Vincent Diepeveen wrote: >On December 30, 2003 at 08:14:20, Mike Byrne wrote: > >>On December 30, 2003 at 03:25:05, Uri Blass wrote: >> >>>On December 29, 2003 at 22:44:08, Mike Byrne wrote: >>> >>>>On December 29, 2003 at 21:49:40, Mike Byrne wrote: >>>> >>>>>On December 29, 2003 at 15:47:05, Vincent Diepeveen wrote: >>>>> >>>>>>On December 29, 2003 at 13:59:55, Mike Byrne wrote: >>>>>> >>>>>> >>>>>> >>>>>>>On December 29, 2003 at 13:23:33, Sandro Necchi wrote: >>>>>>> >>>>>>>>On December 29, 2003 at 12:46:47, Luis Smith wrote: >>>>>>>> >>>>>>>>>>I do agree too. >>>>>>>>>> >>>>>>>>>>Crafty has no realistic chances to win a WCCC. >>>>>>>>>> >>>>>>>>>>Sandro >>>>>>>>> >>>>>>>>>IMO only Bob can know this for sure. I think people either over estimate the >>>>>>>>>commercials, or underestimate Crafty. After all at the WCCC's only 11 games >>>>>>>>>were played, who knows what could have happened in that time, especially with >>>>>>>>>the kind of hardware that Dr. Hyatt could get. >>>>>>>> >>>>>>>>No, Bob does not know this. >>>>>>>>He is a "little outdated" on this matter. >>>>>>>> >>>>>>>>At the 2003 WCCC there were 3 favorites (Shredder, Fritz and Junior), 2 possible >>>>>>>>outsiders (Brutus and Diep). >>>>>>>> >>>>>>>>Based on my experience I gave these chances, before the tournament started: >>>>>>>> >>>>>>>>Shredder 35% (because of the slower hardware) >>>>>>>>Fritz 30% >>>>>>>>Junior 25% >>>>>>>>Brutus 7% >>>>>>>>Diep 3% >>>>>>>>rest 0% >>>>>>> >>>>>>> >>>>>>>The rest of the field is never 0%. Any bookie can tell you that. It might be >>>>>>>15 to 1, 30 to 1 -- even a 100 to 1 . but the chances are never "zero" - that >>>>>>>would make the payoff infinity. >>>>>> >>>>>>Wrong. >>>>> >>>>>Vincent, I love you - you have not lost not one iota of ability to doublespeak. >>>>> >>>>>I'll skip futher down. >>>>> >>>>> >>>>>> >>>>>>When Diepeveen (FM) 2300 plays Jonathan Schaeffer in chess at FIDE level (40 in >>>>>>2 + 20 in 1 + 15, you are correct that he has a 1% chance to win from me. It >>>>>>happens each so many years that i lose from a national master (2000 rated USCF). >>>>>>Statistically his chance is higher by the way than in reality. The only 2000 >>>>>>rated player i lost from last 6 years a 2 hours game (so not fide rated even, >>>>>>only national) was a youth talent who was 2200 rated one rating list later (so >>>>>>underrated). >>>>>> >>>>>>The only reason Jonathan has a chance is because he can play without blundering >>>>>>away all pieces and he knows at which spot a piece is best. He has of course >>>>>>experience playing titled players in tournament games. >>>>>> >>>>>>Depending upon whether he has practiced past weeks, Jonathans chances will be >>>>>>1.5% or 0.5% practically spoken. This is simply not interesting. There is *some* >>>>>>chance. Chance is bigger when i'm feeling a bit sick of course. >>>>>> >>>>>>We can of course argue a long time about how high the chance is and we will >>>>>>never agree i bet. My argument will be he has less than 1% chance because USCF >>>>>>is inflated compared to the european ratings of today. >>>>>> >>>>>>However, >>>>>> >>>>>>Diepeveen - Eric van Reem (1803 national rated in Netherlands) >>>>>> >>>>>>That's a 0% chance for Eric. I will be motivated to my bones to beat someone >>>>>>like Eric of course. >>>>>> >>>>>>Now people will go start using statistics that i might blunder once in my life >>>>>>at move 7 away a piece or something, or that eric has some trick once in his >>>>>>life which he sees and he wins from me. >>>>>> >>>>>>All possible. >>>>>> >>>>>>When i claim to never lose, that's just a claim. At a certain level people >>>>>>simply give away too little pieces to ever lose from very low rated players. >>>>>> >>>>>>But still people will tell here: "well perhaps the chance is 0.0001 but it is >>>>>>possible that once in your life you blunder away that piece against a 1800 >>>>>>rated". >>>>>> >>>>>>This argumentation is true of course. >>>>>> >>>>>> VIRTUAL REALITY >>>>>> >>>>>>But now the reality. I ask the statisticians now: what is the chance that at an >>>>>>11 round match, Eric van Reem(taken many pictures from titled players) will beat >>>>>>Vincent Diepeveen(FM), perfectly healthy and playing for his life, in a 11 round >>>>>>match? >>>>>> >>>>>>But now let's say that i am not so healthy, despite feeling healthy, and by >>>>>>coincidence that week have a virus which kills my possibilities to play well. >>>>>> >>>>>>I bet statistics will say 0.00000000000000000000001 at most now. >>>>>> >>>>>> >>>>>> REALITY >>>>>> >>>>>>The real reality is that in a world champs Eric van Reem isn't only playing FM >>>>>>Vincent Diepeveen. Reality is that everyone is motivated to win. First round he >>>>>>gets GM Alterman, then he gets Omid David Tabibi (didn't play much lately but >>>>>>plays very strong 2200+ hands down), then he gets 2343 FIDE rated Johannes >>>>>>Zwanzger and that for 11 rounds. >>>>>> >>>>>>Now the 0.00000000000000000000001 changes in 0 simply. >>>>>> >>>>>>Imagine next, a 11 round world championship humans. the participant list: >>>>>> >>>>>>1. Kramnik 2777 (note that his matches vs kasparov and such were never >>>>>> counted for FIDE rating, fide has boycotted that. >>>>>> Lucky kasparov) >>>>>>2. Ponomariov 2718 >>>>>>3. Kasparov 2830 >>>>>>4. Anand 2766 >>>>>>5. Adams 2725 ENG >>>>>>5. Svidler 2723 (qualified at internet) >>>>>>6. Polgar 2722 >>>>>>7. Ivanchuk 2710 >>>>>>8. Sokolov, Ivan 2695 NETHERLANDS >>>>>>9. Ye, Jiangchuan 2681 >>>>>>10.Lautier 2666 FRANCE >>>>>>11.Onischuk 2661 USA (highest rated US player who is active) >>>>>>12.Van Wely 2654 NETHERLANDS >>>>>>13.Seirawan 2621 USA >>>>>>14.Bu, Xiangzhi 2606 CHN (Born: 1985-12-10) >>>>>>15.Diepeveen 2276 >>>>>>16.Hyatt 1800 (local rating, FIDE starts at 2000 for international >>>>>> events) >>>>>> >>>>>>time control 40 in 2. rounds = 11 >>>>>> >>>>>>If we play 11 rounds you are now claiming that number 15 and 16 have a tiny >>>>>>chance to win the world title FIDE? >>>>> >>>>>No, show me where I claimed that - classic Vincent doublespeak. In the example I >>>>>used , somebody has picked the top 5 out of 14 and said the rest of the field >>>>>had 0%. To use your example, the rest of the field is with Polgar on down - are >>>>>YOU now claiming that 6 through 16 have no shot. It's not 0.0000000000000001%. >>>>> >>>>>My God, that is biggest nonsense i ever heard! >>>>> >>>>> >>>>>> >>>>>>That is the biggest nonsense i ever heard! >>>>>> >>>>>>The chance is not 0.00000000000000000000000000000001 >>>>>>The chance is 0 exactly. >>>>>> >>>>>>Just like the world champion FIDE 2004 will never be a player rated < 2600 for >>>>>>the very same reason. >>>>>> >>>>>>The only reason bookmakers give 1 to 30, is because they earn more giving 1 to >>>>>>30 than when they would give 1 to 100. >>>>>> >>>>>>>Besides, the tournamnament format , imo, is stupid. An 11 round swiss with 14 >>>>>>>or so participants? - they should make a "normal" swiss (say 5 rounds with 14 >>>>>>>participants or a round robin - add just 2 more rounds with 14 participants. I >>>>>>>forget exactly the number of rounds and participants, but I'm not far off. >>>>>>> >>>>>>> >>>>>>>> >>>>>>>>Do you really believe Crafty is better than the average rest? I do not. He would >>>>>>>>have to rely on too many bugs on the competition. This is not realistic. >>>>>>>> >>>>>>>>Of course everybody can say anything different, but in reality this is the >>>>>>>>situation. >>>>> >>>>> >>>>>The reality is this -- the performance ratings of WCCC at Graz. The odds are >>>>>Crafty would have finished somewehre between List and Brutus according to the >>>>>latest SSDF results. Crafty is just about 200 points behind Shredder on equal >>>>>hardware on the latest SSDF ratings. Your beloved Diep finished nearly 400 >>>>>points behind Shredder in this tournament and would have no shot against Crafty. >>>>> That's reality. >>>>> >>>>>1 Comp Shredder 2734 11 >>>>>2 Comp Deep Fritz 2722 11 >>>>>3 Comp Deep Junior 2632 10 >>>>>4 Comp Brutus 2596 11 >>>>>5 Comp List 2485 8 >>>>>6 Comp Greenlight Chess 2415 10 >>>>>7 Comp Diep 2344 10 >>>>>8 Comp Quark 2323 10 >>>>>9 Comp Chinito 2321 11 >>>>>10 Comp Falcon 2262 10 >>>>>11 Comp ParSOS 2253 11 >>>>>12 Comp Deep Sjeng 2234 11 >>>>>13 Comp Jonny 2228 11 >>>>>14 Comp Nexus 2169 11 >>>>>15 Comp Hossa 1947 10 >>>>>16 Comp Ruy Lopez 1935 10 >>>>> >>>>>note _ calculation was based on actual games played - no forfeit wins - another >>>>>loving dose of reality. >>>>> >>>>>love, >>>>> >>>>>Michael >>>> >>>> >>>>I forgot to mention - the ratings above were derived using the Fritz rating tool >>>>- I made no effort to calibrate them to anything else - they are only good >>>>relative ratings to each other --- not to SSDF, USCF, FIDE etc. It 's the delta >>>>between the programs that may be useful, the absolute values mean nothing in >>>>this context. >>> >>> >>>Based on Leo's rating list >>>see rating list in http://wbec-ridderkerk.nl/ >>> >>>We have on single processor >>> >>>Deep Sjeng1.5 2641(2234 in your list) >>>Crafty19.03 2574(your estimate near 2534 that is shredder minus 200) >>>GreenLightchess3.00 2501(2415 in your list) >>>Quark2.05b 2487(2323 in your list) >>> >>>You seem to reduce 200 from the rating of shredder to calculate crafty's place >>>when you ignore comparison of Crafty with other programs. >>> >>>I think that it was a tournament when the commercial programs were simply lucky >>>so if you give rating based on the tournament the difference between the >>>commercial and the rest should be bigger. >>> >>>Uri >> >>I am not familiar with Leo's rating list -- but in any tournament an engine can >>exceed worse or better than expecations. Perhaps DeepSjeng was unlucky. But >>that just proves the point that we were initially talking about -- where someone >>said the rest of the field (after the top 4 or engines) had no shot and I say >>the rest of the field has a shot. > >a) you misquote >Sandro said: >>>Shredder 35% (because of the slower hardware) >>>Fritz 30% >>>Junior 25% >>>Brutus 7% >>>Diep 3% >>>rest 0% > >Look the problem is you never join tournaments yourself. > >Otherwise you would have made the same table. > >Taking a shot at an engine (winning 1 game) is something different as winning >the event. > >You should not misquote the ELO. Elo is based upon playing 1 game to another and >it is not taking into account draw margins nor facts like that when playing at a >high level chances change. > >In tournaments where also very well prepared professionals join such like events >as this, the chance that some weak program like crafty wins is simply 0%. > >No matter the hardware. > >Bob knows this and that's why he won't show up at serious events anymore. He can >get plenty of operators at any event. Let's be clear about this. > >Graham operated crafty in world champs 2000. >Michel operated crafty in world champs 2001. > >Michel had to make his own book even for that event and already knew he would >not win based upon book games. > >That's the reality of crafty. > >It just has too many weak spots. The strong thing about crafty is it hardly >changes so it is very well tested and Bob *is* very good in tuning. > >But trivially searching deeper doesn't help crafty at all. All those 'crafty >goes deep' experiments were of course a big joke. They just looked to PV changes >not whether a BETTER move came out at bigger search depths. > >Later experiments of Heinz conducted with Fritz, just recently posted in ICGA >clearly show a diminishing return. This despite the fact that the experiments >still do not search at search depths which are considered by me to be above the >tactical barrier. Diminishing returns does not mean that searching deeper does not help. The tactical barrier exists only in your imagination. > >I quoted 12-14 ply always for this. > >Now Fritz has a bigger evaluatoin function than crafty. Frans told it's *a lot* >bigger than what crafty has. The question is about quality and not about size. I doubt if Junior has bigger evaluation than Crafty. > >Don Dailey in selfplay experiments, also at small depths already, clearly showed >diminishing returns. > >All the big tournametns where crafty and you do not show up, show to the >commercial participants clearly that another ply deeper doesn't count when >compared to improving evaluation function. No They prove nothing. I also think that the commercial programs have superior search so it is not competition between search and evaluation. > >In fact the only reason GCP can shout victory now about diep-sjeng is a very >stupid thing my evaluation caused (Qb3??). I could that game not complain about >search depth with 512 cpu's! It only prove that bigger is not better and your evaluation was bigger than the opponents but the opponents had always better evaluation. 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.