Author: Robert Hyatt
Date: 09:01:57 07/18/02
Go up one level in this thread
On July 18, 2002 at 05:41:40, Vincent Diepeveen wrote: >On July 18, 2002 at 00:47:17, Robert Hyatt wrote: > >>On July 17, 2002 at 16:27:48, Amir Ban wrote: >> >>>On July 17, 2002 at 11:44:26, K. Burcham wrote: >>> >>>> >>>>You show below that the two programs play 36.Qb6, not axb5. >>>> >>>>You are actually strengthening Kasparov's question "how could Deep Blue fail to >>>>play 36.Qb6 ?". 5 years later, I've yet to see a program that plays 36.axb5. >>>> >>>>Amir >>>> >>>> >>>> >>>>Deep Blue >>>>Game 2 move 36 >>>> >>>>Fritz 7 on 1533 mhz amd with 432 hash, will play 36.axb5 between 39 minutes >>>> and up to 70 minutes in eval, (see below). >>>> >>>>like I said the three positions after blacks move 40 seem equal. both programs >>>>threaten the same as Deep Blue and will play >>>>without human interference the moves Be4 and axb5. the newer programs accomplish >>>>same objective by playing axb5, just not first move in this line. >>>> >>>>kburcham >>>> >>> >>>It is needed to understand what the fuss was about here. >>> >>>36.Qb6 wins a pawn. Other moves don't. >>> >>>Kasparov expected the natural Qb6 and planned in response to give away two more >>>pawns for an attack, which would have given him some chances. >>> >>>Kasparov did not believe any computer can see the merit of giving up these three >>>pawns. So, when Deep Blue instead played 36.axb5, he asked why. >>> >>>There's nothing special about 36. axb5. If you decide not to take the pawn with >>>Qb6, then this is a natural second choice. >>> >>>If the answer to Kasparov's question is that Deep Blue saw the line Kasparov >>>considered and evaluated it properly, then the question is answered and hooray >>>for Deep Blue. Otherwise it's still unanswered. >>> >>>Amir >> >> >>Their log seems pretty clear to understand here... > >Bob 8(6) means 8 ply fullwidth with at most up to 6 ply >extensions in software for tactics. Like we have check extensions >and other extensions. Vincent, 8(6) means 8 plies of search done on the SP2, 6 _more_ plies done on the chess processors. We have been around this _over_ and _over_ with an email from the DB team posted here for your inspection. They _always_ reported their depth like that. With Deep Thought. Deep Thought 2, because I personally _saw_ those machines in action and I asked Hsu directly. I later asked when we started looking at the DB2 logs from 1997 and they responded that they reported the depths exactly as they had always done so, so that the first number was the depth searched (not counting extensions) and the number in parens was the additional depth tacked on by the chess processors. That "6 plies of extensions" is utter nonsense. Give _any_ reference to justify such a claim. Nothing _I_ have here shows any "6 ply limit." In fact, everything they have written claims that the typical path is extended _far_ more than 6 plies. > >In 1997 i had in DIEP the opposite bishop penalty pretty high. > >Simply say in endgames with rooks and opposite bishop that the >score is 0.00 and that kings without pawns around it and a queen >nearby get 3 pawns penalty, then your thing will play axb5 also >at 7 ply ;) So what? Their scores didn't show any such thing. The log showed that they liked Qb6 less each time they drove the search one ply deeper, until it got so low that axb5 had a higher score and they went with that. > >>at depth 8(6) [14 plies total] they liked Qb6 with a score of 87. At >>depth 9(6) same move, score=79. At depth 10(6) it dropped to 74. At >>depth=11(6) [final search, 17 plies] Qb6 dropped to 48 and was replaced >>by axb5 with a score of 63... >> >>What is hard to understand there? Their search/eval liked axb5 slightly >>more than Qb6. About 1/10th of a pawn better, roughly, assuming they >>still use pawn=128 which I don't know is true for DB2...
This page took 0.03 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.