Author: Dan Homan
Date: 05:55:31 11/18/98
Go up one level in this thread
On November 18, 1998 at 05:00:14, Ernst A. Heinz wrote: >On November 17, 1998 at 06:36:49, Ernst A. Heinz wrote: > >>On November 16, 1998 at 18:19:53, Amir Ban wrote: >> >>>On November 16, 1998 at 14:50:57, Bruce Moreland wrote: >>> >>>>r1r1q1k1/6p1/p2b1p1p/1p1PpP2/PPp5/2P4P/R1B2QP1/R5K1 w - - 0 1 >>>> >>>>It has been contended that this move would be difficult for a computer to find, >>>>and this has caused some doubts to be raised as to whether the computer found it >>>>without assistance in this game. >>>> >>>>I would like to ask how we can clear this up absent input from DB. >>>> >>>>Has anybody run this for a long period on a micro, and if so, was any move >>>>selected other than 35. Qb6? >>>> >>>>Is the counter-attacking line 35. Qb6 Qe7 36. axb5 Rab8 37. Qxa6 e4 supposedly >>>>the reason that white shouldn't play 35. Qb6? Or is it some other line? If it >>>>is too hard or impossible to find 35. axb5, would finding this line show >>>>anything? >>>> >>>>Is there some minimum score delta we can achieve between the position after 35. >>>>axb5 and 35. Qb6 that might be evidence that DB should be given the benefit of >>>>the doubt? >>>> >>>>Are these questions unfair or wrong, if so, are their other questions that can >>>>be asked and possibly answered that will help clear this up? >>>> >>>>bruce >>> >>> >>>This is move 36 but the other details are correct. >>> >>>The first time this was discussed on CCC, Chris Whittington was still here, and >>>he reported for CSTal. It was closer than others, but still couldn't bridge the >>>gap. >> >>I fed this position into the current "DarkThought" yesterday and it liked Qb6 >>with a score of roughly +1.7 up to iteration #18 inclusively. Then, it failed >>low on Qb6 (score <= 1.39) in iteration #19 after processing 6,490,725,565 >>nodes. Currently, it is still engaged in resolving the fail-low (see below). >> >>//////////////////////////////////////////////////////////////////////////////// >> >>r1r1q1k1/6p1/p2b1p1p/1p1PpP2/PPp5/2P4P/R1B2QP1/R5K1 w - - 0 1 >> >># Clearing chess engine. >># Engine sleeps. >>- - >> +------------------------+ >>8 |*R :::*R :::*Q :::*K :::| >>7 |::: ::: ::: *P: | >>6 |*P ::: *B: *P: *P:| >>5 |:::*P ::: P *P: P ::: | >>4 | P :P:*P ::: ::: :::| >>3 |::: :P: ::: ::: P | >>2 | R ::: B ::: :Q: P :::| >>1 |:R: ::: ::: :K: | >> +------------------------+ >> a b c d e f g h >> >>14.01 Qb6 Rd8 Be4 Rac8 Qxa6 bxa4 Qxa4 Qxa4 Rxa4 Rd7 ... (1.71) #62711053 >>15.01 Qb6 Rd8 Be4 Rac8 Qxa6 bxa4 Qxa4 Qh5 Kh2 Qg5 ... (1.69) #140398448 >>16.01 Qb6 Rd8 Be4 Rac8 Qxa6 bxa4 Qxa4 Qh5 Qa7 ... (1.70) #423411650 >>17.01 Qb6 Rd8 Be4 Rac8 Qxa6 bxa4 Qxa4 Qh5 Qa7 Qh4 ... (1.73) #1230726361 >>18.01 Qb6 Rd8 Be4 a5 axb5 axb4 Rxa8 Rxa8 Rxa8 Qxa8 ... (1.64) #3337837066 >>19.01 Qb6 <=178? (1.39) #6490725565 >> >>//////////////////////////////////////////////////////////////////////////////// > >I run this on a 500MHz Alpha-21164a and the fail-low of iteration #19 got >resolved to +1.26 after 14:34 hours and 11,196,023,081 nodes (see below). > >>19.01? Qb6 Rd8 Be4 a5 axb5 axb4 Rxa8 Rxa8 Rxa8 Qxa8 ... (1.26) #11196023081 > >I will let "DarkThought" continue the calculation as long as the other people >at our institute do not get too angry at me for blocking so many Alphas ... :-) > >=Ernst= Does anyone know how long deep blue took on this move? Dark Thought has passed 11 billion nodes which is about where Deep Blue may have been after 1 minute of calculation. (I realize that Deep Blue's search tree and evaluation is vastly different, but this is the only comparision I can make, given my sparse information on Deep Blue's abilities.) I find it interesting that Dark Thought dropped the score of Qb6 by about half a pawn in this time range. Bob suggested in another post that when crafty was more paranoid about king-safety, the score difference between Qb6 and axb5 was less than half a pawn. (correct me if I have remembered incorrectly here) Deep Blue is known (as much as anything about Deep Blue can be known) to have some very strong king-safety evaluation. (I read a post either here or in rgcc that Hsu compared it to CSTal in that respect) It is quite reasonable to me that Dark Thought may have already found the answer here. If Deep Blue found a similar (half-pawn) eval drop on Qb6 after a similar quantity of work (1 minute for Deep Blue), it is quite reasonable to think that eval differences could allow axb5 to come to the top of the list. - Dan
This page took 0.01 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.