Author: Uri Blass
Date: 07:42:04 06/23/01
Go up one level in this thread
On June 23, 2001 at 10:09:44, Robert Hyatt wrote: >On June 23, 2001 at 01:25:40, Uri Blass wrote: > >>On June 22, 2001 at 23:58:45, Robert Hyatt wrote: >> >>>On June 22, 2001 at 17:21:50, Vincent Diepeveen wrote: >>> >>>>On June 22, 2001 at 15:41:33, Robert Hyatt wrote: >>>> >>>>>On June 22, 2001 at 11:26:26, Vincent Diepeveen wrote: >>>>> >>>>>>On June 22, 2001 at 10:50:01, Robert Hyatt wrote: >>>>>> >>>>>>>On June 22, 2001 at 10:37:18, Vincent Diepeveen wrote: >>>>>>> >>>>>>>>On June 22, 2001 at 10:35:10, Harald Faber wrote: >>>>>>>> >>>>>>>>>On June 22, 2001 at 08:28:01, Vincent Diepeveen wrote: >>>>>>>>> >>>>>>>>>>Hello, >>>>>>>>>> >>>>>>>>>>My dual has searched past few days at next position: >>>>>>>>>> >>>>>>>>>[D]2rr2k1/pp1qnppp/2n1p3/3p4/1bPP3P/1P2RNP1/PB3P2/1BRQ2K1 w - - 0 1 >>>>>>>>>>kasparov-deep blue 1996 bxh7!! >>>>>>>>>> >>>>>>>>>>Here is the result of the current diep version (threats on, recaptures >>>>>>>>>>on, checks on, SE on, etcetera on): >>>>>>>>>> >>>>>>>>>>dual PIII800 150mb hash >>>>>>>>>>01:39:58 0 0 415044900 (74138755) 14 0.468 a2-a3 Bb4-a5 b3-b4 Ba5-c7 Qd1-d3 Ne7- >>>>>>>>>>g6 h4-h5 Ng6-f8 h5-h6 g7-g6 b4-b5 Nc6-a5 c4xd5 >>>>>>>>>>++ b1-h7 >>>>>>>>>>04:59:23 0 0 1234506378 (199255362) 14 0.767 Bb1xh7 Kg8xh7 Nf3-g5 Kh7-g6 Qd1-d3 >>>>>>>>>>Ne7-f5 g3-g4 Kg6-f6 g4xf5 e6xf5 a2-a3 Bb4-e7 c4xd5 Qd7xd5 Re3-e1 Be7-d6 Qd3-f3 >>>>>>>>>> >>>>>>>>>>It is busy 42 hours now to get a new mainline for Bxh7 at 17 ply, >>>>>>>>>>so i guess it has seen more than 10 billion nodes now. >>>>>>>>>> >>>>>>>>>>Score is 0.828 for Bxh7 after 17 hours though and my expectation is that >>>>>>>>>>to get the next PV i need to wait another 40 hours at least, and that the >>>>>>>>>>score hasn't changed too much, and i need my dual now to do other things! >>>>>>>>> >>>>>>>>> >>>>>>>>>So you want to say that you just need a 50x faster machine (or 100x to be safe) >>>>>>>>>that your Diep plays like Kasparow (:-)))) and makes the move in a tournament >>>>>>>>>game? :-) >>>>>>>> >>>>>>>>Deep Blue's score of 0.00 was not correct. >>>>>>>> >>>>>>>>Note that a3 c5 b4 also wins chanceless for white, as proven by Kasparov >>>>>>>>and obvious for any chessplayer. >>>>>>>> >>>>>>>>Nevertheless Bxh7 is a cool shot for white. >>>>>>> >>>>>>>Note that DB's score was a hell of a lot more correct than yours. IE when it >>>>>>>played its last move, it _expected_ Bxh7 and had seen a repetition. Whether the >>>>>>>draw was forced or not, it saw it after a couple of minutes from the move before >>>>>>>where you are searching. How long did it take you to get an idea Bxh7 was >>>>>>>good? Just how well do you compare to DB? Back up one move and let your >>>>>>>program search until it sees Bxh7 for your _opponent_. Then you will begin to >>>>>>>see how deeply they could search. >>>>>>> >>>>>>>You are at the wrong position and it _still_ takes you forever to find Bxh7. >>>>>> >>>>>>This is utterly nonsense, >>>>>> a) diep doesn't play the same horrible moves which DB played >>>>> >>>>> >>>>>Did you read what I wrote? I said tht you are trying to find Bxh7 and having >>>>>a _hard_ time doing it. DB found it for the _opponent_ as it expected this in >>>>>the game. This means that after you strip off one ply of its search, it _still_ >>>>>cound see the tactics that make Bxh7 look interesting. You can not find it in >>>>>2-3 minutes from Kasparov's move... yet DB found it when it was trying to choose >>>>>a move. >>>>> >>>>>Get it now? >>>> >>>>You don't want to get it, but in order to get this lost position, >>>>there DB played somehorrible moves which diep doesn't play because >>>>it knows a bit more! >>> >>>Here is something that convinces me it does not. >>> >>>Deep Thought beat Cray Blitz every time we played them. >> >>I understood that there was only one game of Cray blitz with no bugs against >>Deep thought and if you decide based on one game then >>Fritz3 also beated Deep thought. > >I disagree there. I would not _ever_ call a "parallel search program" as >a "program with no bugs". I don't believe we _ever_ played a game where >I wasn't "holding my breath". I am _certain_ that Crafty still has its >share of quirks and flaws, in fact... > > > > > >> >> >>> >>>I played a 20 game match vs Crafty a few months back and posted the results. >>>I don't remember now, but it was something like 13-7 in favor of CB. CB >>>used a T90, Crafty used my quad 700. >>> >>>Crafty wins more against Diep than it loses on ICC. >>> >>>I don't see therefore how you could even say Diep knows way more than >>>Deep Thought, much less Deep Blue or Deep Blue 2. >>> >>>I have watched your program play horrible move after horrible move. Mine does >>>too. You imply that yours does not, but that is hogwash. Particularly in an >>>endgame DB, Kasparov, or many other programs will wreck you quickly.... >>> >>> >>> >>> >>> >>>> >>>>> >>>>> >>>>> >>>>>> b) the reason diep doesn't initially play Bxh7 is because it >>>>>> wants to win the game by means of pawn majority, just like kasparov, >>>>>> only when it sees tactical way more as DB it plays Bxh7 >>>>> >>>>>That is funny. DB saw this as the second move in its PV from the previous >>>>>search. you can barely find it as the _first_ move in your PV and it takes >>>>>forever to do so. And you see "tactically way more as DB"??? >>>>> >>>>>I don't follow at _all_.. >>>> >>>>Score is wrong from DB. We can discuss in length, but that's it. >>>>My score is better when i get Bxh7, with DB that would take forever >>>>to get 14 ply with SE and fullwidth. >>> >>>14 plies doesn't take them forever. They hit 14 _all_ the time, just look at >>>the logs. 14 is actually _low_. They are generally deeper than that. >> >>I see no reason to argue because >>It is clear there is a disagreement if 11(6) mean 17 plies or 11 plies, >> >>Vincent believes that it means 11 plies. >> >>I do not know but based on some analysis my impression is that 11(6) of Deep >>blue is eqvivalent to less than 17 plies of top programs. >> >>My impression is based on the fact that top programs can see similiar main >>lines that Deeper blue saw at 11(6) at depth 15 or 16 when these top programs >>include Crafty. > >Just don't forget that you can't see _all_ of the "main lines". The last 6 >plies plus extensions plus q-search are not accessible due to the chess >processor design. > > > > >> >>I may change my mind if you show me cases when top programs need depth 19 or 20 >>to see things that Deeper blue saw at depth 11(6). > > >Did any program ever decide to give up on the move in game 2 (I don't remember >whether it was axb5 or Qb6... but it was a move questioned by Kasparov. DB >went into "panic time" because the eval dropped way down and it changed to an >unexpected move. That was one possible position that could use analysis. I want to see cases when programs can see the right move after a long search. If they cannot see even after many hours then I suspect that it is not the right move. > > > > >> >>It will be interesting if you can find positions when Crafty needs depth 19 or >>20 to see the same beginning of the main line that Deeper blue suggested at >>depth 11(6). >> >>Uri > > >there was one such position I remembered in match one. Everyone suggested one >rook move, but a different move was very important. Unfortunately, that is >something I don't recall with any degree of detail as it wasn't very important >at the time. I remember it and I remember that Amir ban posted analysis of Junior6 and it could find the right rook move in a few miinutes. Dark thought had probably worse evaluation so it needed some hours to find the same move. At the time of the match programs could not find the right move but I suspect that they got better. I also do not know the search depth of deep blue in the first match so I prefer positions from the second match. 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.