Author: Vincent Diepeveen
Date: 09:23:52 02/26/02
Go up one level in this thread
On February 25, 2002 at 15:11:09, Slater Wold wrote: >On February 25, 2002 at 12:55:37, Vincent Diepeveen wrote: > >>On February 25, 2002 at 11:46:21, Robert Hyatt wrote: >> >>>On February 25, 2002 at 11:35:02, K. Burcham wrote: >>> >>>> >>>> >>>>I agree, but it seems the problem with playing on the Deep Blue level is >>>>kns. If we are getting 2000kns today, with the dual machines, then we will >>>>have to wait on hardware upgrades. Hyatt said that Deep Blue was getting >>>>200,000,000 nodes a second--we are only getting 2,000,000 nodes a second. >>>>Hardware will have to reach closer to this level to test todays search methods >>>>in comparison to Deep Blue's search methods. If this is true what Hyatt is >>>>saying, and I have no reason not to believe this, then I have learned my 3100 >>>>mhz and 1000 megs ram, is very limited in being able to compare todays programs >>>>to the six game Deep Blue match with Kasparov. The more time I spend testing >>>>positions in this match, the more I dissagree with Uri. I feel if Uri had >>>>spent the time and money that I have in testing this match, then I dont think he >>>>would take such a stand comparing todays programs with Deep Blue. >>>> >>>>I have started over again with the six game Deep Blue match. I am using the >>>>following: >>>> >>>>Fritz7 1500mhz 512ram >>>>Chess Tiger 14 1000mhz 384ram >>>>Deep Shredder6 3100mhz 1000ram >>>> >>>>Starting with game one, I will run all three programs until a Deep Blue move >>>>is not played by any program. I will then give all three programs 24 hours to >>>>find the Deep Blue move. If none of the programs will play the Deep Blue move >>>>after 24 hours, I will move to the next move. This move will be noted. >>> >>>This is going to be a tough thing to evaluate. IE (a) is the DB move better? >>>(b) is the DB move worse? (c) Is the alternative suggested by one of the above >>>just as good? >> >>Indeed using programs is no good idea to be sure of things. >>Seirawan has written real good analysis about the DB match. There >>is just 1 move where i disagree with his comments (from chess technical >>viewpoint). The game 2: Qf1 move which Seirawan gives a ?! mark and giving Qf2 >>as alternative. Qf2 is losing to a simple queen capturing tactic. >> >>Also GM v/d Wiel had missed this tactic, but he hadn't prepared very >>well for his lecture on the match. >> >>Nevertheless, Seirawan wrote loads of pages with excellent comments and >>many positions where the programs made a real bad move. >> >>I do believe that a programs strength *can* be shown bad by showing how >>many very bad moves it is playing. >> >>Let's not confuse the modern 'shuffle' move of computers with strategic or >>positional blunders. A shuffle moves quality is hard to measure, whereas >>strategic/positional blunders are clear moves which must get avoided. >> >>the number of nodes a second DB got is irrelevant here as it searched >>in a different (outdated) way. Best is to compare plydepths that >>programs get after a minute or 3 to 5. > > >Not exactly. 16 ply to DB is not 16 ply to DF. DF uses null move, and DB did >not. 16 ply to DB is much deeper than 16 ply to DF. first of all it only got 11 to 12 ply. Secondly it is not deeper at all. it is the same depth. It only means sometimes you *might* prune away something which DF sees at 12 which DB sees at 11. In principle minimax nor nullmove nor alfabeta nor hashtables give for a depth n the same score back to the root. Nevertheless *no one* is using minimax without alfabeta pruning. Nowadays nearly no one is not using hashtables+nullmove. >>>It is like trying to pick a particularly bad move by RC4 in the van Wely >>>match, and asking "which computer won't play this move?" That is totally >>>pointless, unless, in addition, you confirm that the computer _will_ play all >>>the preceeding moves so that it would actually reach that position and then >>>improve on what Rebel played. It might be that there are better moves to be >>>played earlier in the game. It might be that Rebel's play up until the "lemon" >>>move was flawless. >>> >>>Proving any of that is a serious undertaking... >>> >>> >>> >>> >>>> >>>>I wanted to use Century 4.0, but I have not learned all of the interface >>>>necessary to cut and paste, clipboard, save as, etc. with the windows-dos. >>>> >>>>I feel game one will go very fast, because I think most moves can be found in >>>>this game. It seems some settings were changed between game one and game two, >>>>because finding moves after game one became more difficult. >>>> >>>>Anyone have any advice on how to improve this project, please let me know. >>>>I will record: 1. moves not found 2. moves found after 30+ minute search >>>>3. which program found which moves. 4. Total time of analysis after each game. >>>>kburcham
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.