Author: Uri Blass
Date: 02:07:10 12/03/02
Go up one level in this thread
On December 03, 2002 at 01:53:22, Les Fernandez wrote: >On December 03, 2002 at 01:24:57, Uri Blass wrote: > >>On December 03, 2002 at 00:49:35, Les Fernandez wrote: >> >>>On December 02, 2002 at 16:41:16, Uri Blass wrote: >>> >>>>On December 02, 2002 at 16:14:35, Les Fernandez wrote: >>>> >>>>>As hardware speed increases, better algorithms for pruning and faster search >>>>>routines get better we will start to see some of this imagination all of us >>>>>would love to see now. As search depths go deeper more and more positional >>>>>effects will be realized since positional play is related to how deep and >>>>>thorough your search is. >>>>> >>>>>Les >>>> >>>>No >>>> >>>>The problem is not that the program does not play the best move but that the >>>>program does not play the move that is considered by the player as better. >>> >>>Hi Uri, >>> >>>I think I misinterpreted the imagination part. To me it meant that a chess >>>engine could produce a series of moves, that had it been a human player, would >>>have been considered fascinating and ingenius. >>>> >>>>After it prefers another winning move the player starts to complain that >>>>computer have no imagination. >>> >>>Once computers can search deep enough I dont believe that the player can even >>>understand the complexity of the decided move. >> >>I believe that it already happens. >> >> >> I agree that perhaps due to the >>>player not understanding the move the player may consider it unimaginative but >>>keep in mind that it may be past the horizon of the human player. Through the >>>years there have been many grandmasters whose play was considered stale and >>>boring and we never gave that much thought. There will come a point (50,100 or >>>200 years) where computers will be able to just out depth humans. The bottom >>>line is, whether its boring or exciting, that its not how it plays the game its >>>the results that count. >>> >>>Just my .02 worth. >>> >>>ps how are you making out with your chess engine? >> >>I plan to continue to make progress by better search rules. >>I believe that better pruning rules can help movei to be significantly faster. >> >>I believe that programs search too big trees and it is possible to reduce them >>by at least 90% if you make the right observations. > >Wow ! 90%? Uri do you think that is obtainable? I believe that it is obtainable. The only question is if I am able to obtain it. I am optimistic and I believe that I have good ideas and my main disadvantage is programming. I believe that being in average 10 times faster in tactics is not an impossible task. Pruning rules may miss sometimes tactical ideas at the same depth but when I say trees that are 90% smaller then I mean 90% smaller to get the same task and not 90% smaller to get the same nominal depth. I believe that pruning rules that are not null move pruning can make it. That would be a most significant >contribution to chess research if that can be done. Lots of luck and please >keep me posted occassionally as to how you are progressing. Thanks but I guess that it will be contribution to chess research only if I publish my ideas. I believe that movei already use some productive ideas but they help it only to be 2 times faster at long timeand not 10 times faster and movei still does not use a lot of good ideas about extensions and does not use hash tables in an effective way and has no assembler optimization so the general result is that movei is still weaker in tactics relative to the top amateurs and knightdreamer is still clearly faster than movei in the GCP test suite(more than 6 times faster). Here is some comparison between knightdreamer and movei Knightdreamer on PIII733 Mhz 0.1 29/183 1 76/183 5 118/183 20 140/183 Movei on AMD 1000Mhz 5 76/183 21 118/183 116 140/183 You can see that knightdreamer is about 5 times faster than Movei inspite of inferior hardware. I do not know the results of Ruffian but I can imagine that the gap is even bigger. 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.