Author: Joseph Merolle
Date: 07:54:22 03/27/05
Go up one level in this thread
On March 27, 2005 at 09:47:31, Rex wrote: >On March 27, 2005 at 09:03:52, William Penn wrote: > >>I believe that I have just had a new insight here. I've run hundreds of these >>tests with different positions. It looks like .pl2 remembers prior analysis OK >>when the best response involves a capture. Otherwise it forgets, or doesn't >>access .pl2 properly. >> >>For example: >>The final position in a correspondence game was analyzed in infinite mode for >>over 8 hours. The best continuation was found to be 35.Red2 Nc4 36.Bxc4 bxc4 >>with an evaluation of -0.04. So I sent my move 35.Red2 to opponent, and opponent >>replied with 35...Nc4. That gave the following position: >> >>6r1/1k1r1pp1/p1b1p2p/1p6/1PnB4/1BP2P2/P2R2PP/2KR4 w - - >> >>Now, when reloading the game and starting analysis of this position, it >>immediately remembers that the best continuation is 36.Bxc4 bxc4 and assigns the >>-0.04 evaluation value obtained in the long 8+ hour analysis above. This is >>obvious as it goes through the cycles to higher and higher ply levels in >>infinite mode. The first move pair 36.Bxc4 bxc4 and the -0.04 evaluation is made >>instantly as each new ply level begins its calculation. It only remembers the >>best first move pair, so must recalculate all of the alternate move >>possibilities, nevertheless it is saving a lot of time in this new analysis. The >>prior 8+ hour analysis reached a level of 27/47 after 8+ hours. This new >>analysis (in current progress) has achieved the 23/23 level in 5 minutes, the >>24/24 level in 10 minutes, and the 25/25 level in 19 minutes. So it is saving a >>lot of time. >> >>Too bad that this only happens when a capture is involved, but not otherwise. >> :o( >>The reason for this strange behavior is a big mystery still to be solved. I can >>only guess, but suspect a coding error related to accessing the .pl2 file. >>WP > > >I do agree with this. I noticed the same when it learns a position from within >one gui then open Shredder up inside another gui and throw the engine same >position. It will find the correct move in same amount of time. When closing >shredder and reopening shredder is when it doesnt remember learned position. I think I have found the solution to the problems guys. When you analize a position you can not interute its thinking untill it moves. For example you load a position and only let it think for 1 min then you interute or force it to move it will learn the position to that point, but later if you reload it and give it 1 hour to think it will obiviosly find the better move however if you close out and reload it , it will pick up the analise from the one min analyse not the one hour. It seem s learn the position only from its first go around of analysis unless there is a mate involved. I guess this is because it might think there is even a better move to be found that the 1 hour analsis could not find , and this vary well could be posible.
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.