Author: Robert Hyatt
Date: 14:53:31 05/28/02
Go up one level in this thread
On May 28, 2002 at 13:17:57, Albert Silver wrote: >On May 27, 2002 at 14:02:47, Uri Blass wrote: > > >I don't understand the purpose of changing the first move because of a loss. >This implies that the first move was the cause of the loss, when it could easily >(probably) been due to an error much later. If I play a wrong move in an Evans >Gambit and lose because I didn't see far enough, does that mean the fault was >1.e4? > >If the plan is to learn from previous moves, I would imagine that if a time >control and the conditions (memory + CPU, etc.) is the same then it should play >the same first move instantly, presuming that identical conditions will produce >identical result, and then begin calculating again. The actual learning, to >rpevent repetition, should logically appear just before the move where the >evaluation dropped. Perhaps the program could be requested to begin studying the >first two moves prior to the drop in greater depth. Example, if the eval dropped >at move 6, then begin calculating at move 4 the same move, checking the log for >the last depth achieved, and requesting the program go one ply deeper to see if >it can detect the error. If no changes are found, then when it reaches the move >prior to the eval drop, it will of course exclude the losing move though >retaining the evaluation. Of course the request must respect the time control >algorithms. Finally, the reason the first move alone should be played instantly, >and not all the first moves before the perceived error, is that once it plays >one move instantly, the time left changes, and this means that the next move may >change now that it has more time to spend on its remaining moves. > >I should point out that I have no real knowledge of how learning is done >nowadays, so it is essentially speculation. > > Albert That is essentially how I do it. But depending on how bad things look after leaving book, it is possible that not only the last book move (where a choice was available) will be flagged as bad, but it might propogate back up to the previous move where there was a choice also... > >>>Please make a difference between position learning and >>>booklearning. A program that's doing booklearning might >>>be brilliant, but it will keep on losing the same game >> >>In a match of 40 games there is no chance to lose the same game twice if >>you do simple learning that is only to change the first move after a loss >> >>If every first move is a book move then book learning is enough. >> >>The book may have some priorities so 1.e4 f6 is going to be played only if in >>the last 19 games with black you lost with all the alternatives. >> >>You can lose eqvivalent game(1.e4 e6 2.d4 and 1.d4 e6 2.e4) and in order to >>avoid this, a positional learning is needed. >> >>This means that if you get the same position that you lost twice you have to >>play a different move and the simplest way that I can think is that the program >>that lost with 1.e4 e6 2.d4 d5 may assume that 2...d5 is illegal after 1.d4 e6 >>2.e4. >> >>The program may safely forget everything that is not in the last 20 games with >>the same color because the opponent should be a different opponent after 40 >>games(it is also possible to do it after 21 games or 22 games with the same >>color to prevent problems and the only problem is in case that the program lost >>with all the alternatives with the same color in the first 20 games but it is >>not going to happen with one of the top programs. >> >>My point is that I guess that the top programs may get rating that is not more >>than 100 elo weaker than the rating that they got in the ssdf and in some cases >>they may beat again and again programs that has no position learning by moves >>like 1.h3 when the version with the fritz book is not going to do it because it >>is not going to get the opponent out of book in the first moves. >> >>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.