Author: Ed Schröder
Date: 13:38:18 07/19/00
Go up one level in this thread
On July 19, 2000 at 14:12:14, Robert Hyatt wrote: >On July 19, 2000 at 12:53:59, Ed Schröder wrote: > >>On July 19, 2000 at 12:04:25, Robert Hyatt wrote: >> >>>On July 19, 2000 at 00:51:49, Ed Schröder wrote: >>> >>>>On July 18, 2000 at 21:58:45, Robert Hyatt wrote: >>>> >>>>>On July 18, 2000 at 16:49:19, Peter Kappler wrote: >>>>> >>>>>>On July 18, 2000 at 16:03:28, Ed Schröder wrote: >>>>>> >>>>>>>On July 18, 2000 at 14:08:54, Robert Hyatt wrote: >>>>>>> >>>>>>>>On July 18, 2000 at 11:38:01, Ed Schröder wrote: >>>>>>>> >>>>>>>>>On July 18, 2000 at 10:58:51, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>>On July 18, 2000 at 02:00:31, Ed Schröder wrote: >>>>>>>>>> >>>>>>>>>>>On July 17, 2000 at 20:08:06, Robert Hyatt wrote: >>>>>>>>>>> >>>>>>>>>>>>On July 17, 2000 at 17:02:22, Peter Kappler wrote: >>>>>>>>>>>> >>>>>>>>>>>>>On July 17, 2000 at 16:09:09, Amir Ban wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>>On July 17, 2000 at 07:22:41, Graham Laight wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>>>I'm afraid I still feel that Junior could have come out ahead (instead of >>>>>>>>>>>>>>>level)in this tournament by beating Bareev and Khalifman - and possibly by not >>>>>>>>>>>>>>>losing with such apparent ease to Kramnik. Continuing the game against Anand >>>>>>>>>>>>>>>might possibly have gained an extra half point as well. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>I think that Amir has an aspiration to make his program demonstably better than >>>>>>>>>>>>>>>Deep Blue (this certainly comes across in his interviews published on the >>>>>>>>>>>>>>>Chessbase Website coverage of Dortmund (www.chessbase.com) before the Kramnik >>>>>>>>>>>>>>>game). If so, as a (hopefully!) impartial member of the viewing public, I'm >>>>>>>>>>>>>>>afraid to say that I've yet to be convinced. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>As evidence, I point firstly to the games against Bareev and Khalifman. On both >>>>>>>>>>>>>>>occasions when Deep Blue '97 gained an advantage over Gary Kasparov (who's a >>>>>>>>>>>>>>>better player than anyone at Dortmund was), it parlayed that advantage into >>>>>>>>>>>>>>>victory - whilst Deep Junior twice failed conspicuously to "slam in the lamb". >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>I would also point to the game against Khalifman. Here we see Deep Junior lose >>>>>>>>>>>>>>>to a combination of blocked centre and king attack - classic anti computer >>>>>>>>>>>>>>>methods which have both been well known for a long time. They work because, in >>>>>>>>>>>>>>>this case, nothing short of truly massive search depth is going to help you to >>>>>>>>>>>>>>>make the correct moves. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>However, for both king attack and blocked centre, Deep Blue '97 demonstrated >>>>>>>>>>>>>>>that it's evaluation knowledge was able to adequately handle the challenge. >>>>>>>>>>>>>>>Indeed, in game 2 in '97, Deep Blue not only handled the blocked centre, it >>>>>>>>>>>>>>>turned it into a win! >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>It took Deep Blue 2 attempts to beat Gary Kasparov, the world's best player - >>>>>>>>>>>>>>>maybe another year of work will push Deep Junior to a position where it can try >>>>>>>>>>>>>>>to win these tournaments, instead of settling for a middling position. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>But let's not be completely churlish - Dortmund 2000 was indeed a fantastic >>>>>>>>>>>>>>>performance by Deep Junior - and a landmark in computer chess history, since >>>>>>>>>>>>>>>here is both a computer and a program which one can buy in the shops! >>>>>>>>>>>>>> >>>>>>>>>>>>>>I disagree with most of this, but it's your opinion, and if experience teaches >>>>>>>>>>>>>>us anything, it's useless to argue. >>>>>>>>>>>>>> >>>>>>>>>>>>>>For the record, I'm not trying to prove that I'm better than Deep Blue. I think >>>>>>>>>>>>>>I've already shown this some time ago, and I'm not the only one who can say so >>>>>>>>>>>>>>either. >>>>>>>>>>>>>> >>>>>>>>>>>>>>Looking at the (very few) games of DB, I don't see that it had either better >>>>>>>>>>>>>>evaluation or deeper search than today's top programs. >>>>>>>>>>>>>> >>>>>>>>>>>>>>Amir >>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>>I must say I'm skeptical, though I would have a good laugh if it were true. >>>>>>>>>>>>> >>>>>>>>>>>>>Are you aware of any positions from the 2nd Kasparov-DB match where Junior (or >>>>>>>>>>>>>any other micro) plays a clearly better move than DB? Not that this would >>>>>>>>>>>>>conclusively prove a thing - it would just be interesting. >>>>>>>>>>>>> >>>>>>>>>>>>>--Peter >>>>>>>>>>>> >>>>>>>>>>>>Or we can take a few of the positions from the DB log files and try them on >>>>>>>>>>>>"top programs". I'm not aware of any "top program" that can do 16-18 plies >>>>>>>>>>>>in the middlegame... >>>>>>>>>>> >>>>>>>>>>>Please post these positions that would be fun and you might be surprised >>>>>>>>>>>about the outcome. But the key-moves must be clear as there should be no >>>>>>>>>>>discussion what is the best move. I for example don't believe the Rc6 vs >>>>>>>>>>>Rc7 is a good position as this is a case of 0.10 (or so) in evaluation. >>>>>>>>>>> >>>>>>>>>>>Ed >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>I am not talking about "key move" at all. I am talking about doing a full-width >>>>>>>>>>exhaustive search to depth 16-18 in the middlegame, in the same positions where >>>>>>>>>>DB did 16-18 ply full-width searches. >>>>>>>>>> >>>>>>>>>>I don't believe _anybody_ can match their depth/speed. >>>>>>>>> >>>>>>>>>All fine but where are the promised positions from the log-files... >>>>>>>>> >>>>>>>>>Ed >>>>>>>> >>>>>>>> >>>>>>>>I didn't "promise" any positions: >>>>>>> >>>>>>>Yes you are right after I read again. I thought you had some challenging >>>>>>>positions for us poor micro users to compare. What a pity. >>>>>>> >>>>>>>Ed >>>>>>> >>>>>>> >>>>>> >>>>>>I would suggest that we take a close look at some of the positions from game 1 >>>>>>of the '97 match. It's the only game that DB lost. Could the micros have >>>>>>avoided some of those mistakes at tournament time controls? >>>>>> >>>>>>BTW, I don't think it's interesting to compare search depths with the IBM >>>>>>logfiles, as Bob has suggested. I think Amir's original point was that Junior >>>>>>searches as deeply as DB in the lines where it matters. >>>>>> >>>>>>--Peter >>>>> >>>>> >>>>>I would claim that point is nonsense. You only have to read their papers on >>>>>DT and DB to see what their search was doing, extension-wise. It was quite >>>>>sophisticated, as it proved over and over against micro programs in ACM events. >>>> >>>>I would claim that point is nonsense. Chess is about playing the right >>>>moves. We have seen machines with >200 processors losing all the time >>>>from the micros. If memory serves me well no multi-processor system was >>>>able to win the world champion title since 1992, Deep Thought included. >>> >>>And? Let's try the following dates: >>> >>>1983, 1986 (both cray machines) >>> >>>1989, special purpose machine >>> >>>1992, no "big iron" present (cray blitz, deep thought). >>> >>>1995 DT lost one game. >>> >>>So your statement, while true, is not exactly revealing of what went on. My >>>program is not doing badly today. It is (except for wild null move R) very >>>similar to Cray Blitz of 1995 in terms of search extensions and knowledge >>>(except I am not yet using singular extensions as I did in 1995 CB). I don't >>>think the micros were as far ahead of the 'big iron' as you want to believe. >>>You think it was all hardware. It wasn't. And I agree that it still isn't >>>today. But a better program, on faster hardware, will beat a good program on >>>slow hardware most (but not all) of the time. >> >>We have been going through this issue how many times? :) >> >>Fact is this is the year 2000 five years after Hong Kong where everybody >>included me expected DT to become the new world champion due to its huge >>hardware advantage. It did not happen. >> >>I don't see any reason why this could not happen again. >> >> > > >I don't either. But take the following scenario: Someone walks up to you and >says that all the programs in the world are going to gather, including the 97 >version of deep blue, and they are going to play a round-robin chess tournament. >You have to pick one program to win the event. If you are wrong, you lose your >head. You _must_ bet on one program to win, you have no other choice. > >Who do _you_ bet on? To me it is a no-brainer... > >I wouldn't be happy having to bet on DB, but I can't think of anyone that would >have a better chance to win, even though we _know_ that "crap happens" and any >program can lose a game under the right circumstances. > >But what about probability? I think it would be open and shut to pick the >program with the best chance, with no thought at all required. > >And yes, there would be a significant chance that you will lose your head. But >if you pick anybody _but_ DB, the probability goes up dramatically that you will >one day star in "The Legend of Sleepy Hollow". :) I have a more realistic scenario. Micro's have made tremendous progress the last years. Imagine that all the programs which participarted last year in Paderborn would meet again including DB. Micro's will enter with more multi-processor systems than last year. In that extremely strong field it's not unlikely DB will lose one or two games + a couple of draws. Okay, worst scenario but quite well possible. If this happens your end up on place 5-7 or so. Like last year in Paderborn not always the strongest hardware wins. A simple Pentium 550 ended number 1 despite of the many multi-processors around. Faster hardware makes you a big favorite but that doesn't mean you will win as Hong Kong has proven so convincingly. Ed >>> >>>>Chess isn't about ply-depth only. One little mistake and a game can be >>>>over no matter how many processors are added. Chess is about playing the >>>>right moves. >>> >>> >>>Sure... but in many cases "depth" reveals the right move where "knowledge" >>>doesn't. And vice-versa of course. >>> >>>> >>>>So it would be very interesting to have a set of critical positions (not >>>>necessarily mistakes) from the DB-GK matches do a comparison. >>>> >>>>Ed >>> >>> >>>I am trying to go thru the logs to see if I can spot any. The logs are >>>'condensed' which makes parsing (by me) very slow and painstaking... >> >>I know and I hoped you would take the lead here :) >> >>I have been thinking to write a utility that reads a cleaned log-file and >>creates from each DB move an EPD record including the depth, score and main >>variation. hint... hint... wink... wink... > >) > > >I have such a utility for my output logs. But I clearly mark fail high/low >conditions as they happen. In deep blue, for a single iteration, you often >get just one line of output, with moves, scores, times, hash marks (#) and >other stuff crammed together. Fail lows are not easy to spot as you have to >find the best score on the previous iteration and compare it to the first >score for the current iteration. > > > > > >>It would be nice to proof the DB superiority above nowadays micro's and >>close the subject once and for all. >> >>Ed > > > >I am looking as I have time... but it is slow going...
This page took 0.08 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.