Author: Robert Hyatt
Date: 09:06:22 02/04/02
Go up one level in this thread
On February 03, 2002 at 20:21:04, Jeremiah Penery wrote: >On February 03, 2002 at 19:57:59, Robert Hyatt wrote: > >>On February 03, 2002 at 13:43:28, Jeremiah Penery wrote: >> >>>On February 03, 2002 at 07:40:27, Albert Silver wrote: >>> >>>>I'm a bit fuzzy on the accuracy of the PV we're seeing in the logs then. >>>>Presuming that the PVs are only the software PVs, then these may still have been >>>>subject to changes afterwards, no? After all, it's not uncommon to search to a >>>>given depth and say that move A is best, but with greater depth (as the hardware >>>>will provide) move B is shown to be best. I haven't examined the logs in detail >>>>as some here have, so I'm presuming that such an inconsistency isn't there, but >>>>if the hardware extensions aren't capable of changing the decision making, what >>>>good are they? Or were they simply fortunate that this never happened in the >>>>match and that is why we don't see it in the logs. I.e. a main move in the >>>>software-based PV that was different from the move actually played due to later >>>>corrections provided by the hardware extensions. >>> >>>It seems this is what happened on the famous move 36 of game 2, where it >>>suddenly changed its mind from Qb6 to axb5. I don't know if it happened >>>elsewhere. >> >>Not at all, there. It just used way more time to find a new best move as >>the previous best move had dropped in score significantly. > > >This is the relevant portion of the log: > >--> 35. Bxd6 <-- 5/46:59 >--------------------------------------- >hash guess Bf8d6b,Guessing Bxd6 > 8(4) #[Qb6](30)[Qb6](30) 30^ T=1 >qf2b6 Qe8e7 pa4b5P Ra8b8 qb6a6P Pe5e4 bc2e4P Qe7e5 pg2g4 Rc8e8 be4g2 Qe5h2 kg1f1 >Re8e3 qa6c6 Re3c3p > 8(6) #[Qb6](61)[Qb6](61) 61^ T=1 >qf2b6 Qe8e7 pa4b5P Ra8b8 qb6a6P Pe5e4 bc2e4P Qe7e5 pg2g4 Rc8e8 be4g2 Qe5h2 kg1f1 >Re8e3 qa6c6 Re3c3p > 8(6) #[Qb6](87)################################ 87 T=5 >qf2b6 Qe8e7 pa4b5P Ra8b8 qb6a6P Pe5e4 bc2e4P Qe7e5 pg2g4 Rc8e8 be4g2 Qe5h2 kg1f1 >Re8e3 > 9(6) #[Qb6](79)################################ 79 T=18 >qf2b6 Qe8e7 pa4b5P Ra8b8 qb6a6P Pe5e4 bc2e4P Qe7e5 pg2g4 Rc8e8 be4g2 Qe5h2 kg1f1 >Re8e3 >10(6)<ch> 'Bd6' > #[Qb6](74)################################ 74 T=82 >qf2b6 Qe8e7 pa4b5P Ra8b8 qb6a6P Pe5e4 bc2e4P Qe7e5 be4f3 Rc8d8 qa6a7 Qe5c3p >bf3h5 >11(6)[PT=551]?[4 sec (main.c:1391)] #[et1 551 sec][Qb6](48)#[axb5](63) 63 T=416 >pa4b5P Pa6b5p qf2b6 Ra8a2r ra1a2R Bd6c7 qb6e6 Kg8h8 bc2e4 Rc8b8 ra2a6 Qe8d8 >pd5d6 Bc7b6 >--------------------------------------- >--> 36. axb5 <-- 4/40:28 > > >Qb6 dropped in score from 74 to 48. If they used pawn=128, that's like 1/5 of a >pawn - I would hardly call that a "significant" drop in score, worthy of more >than doubling the normal time limit. But you may be right, in that this is a >case of "normally" finding a better move. IMO, it's a bit difficult to tell >exactly what's going on in their logs sometimes. I disagree. Crafty considers .24 a _significant_ drop and will more than double the time limit it established, in order to avoid such a drop. If the score drops .67 then it will go even farther. And once real material goes, it can go up to 6x or something... And yes, their logs were apparently never though of in terms of public consumption. They are hard to read. I made mine pretty clean in the 1970's because we always had folks looking over our shoulder and I got tired of explaining what the output meant. :)
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.