Author: Mogens Larsen
Date: 11:10:41 05/28/02
Go up one level in this thread
On May 28, 2002 at 13:17:57, Albert Silver 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? It depends on the level of sophistication. Just changing or keeping the first move is more in the vains of ensuring diversity than actual learning. The exception being the attempt to repeat a won game. As you mention, book learning makes sense moving backwards from the last bookmove. The most logical IMO approach is to go to the last branching point, ie. the point where there is more than one playable choice. Then there are two general options IMO. Either cut the branch that lead to the defeat or decrease its value and subsequent moves by a (decreasing) factor depending on gameconditions. In the latter case it might be beneficial to go back another branching point. When learning from a positive result, the latter option means an increase in value. The branch cutting option leaves (:-)) everything unchanged in that case, because there's no reason to cut the other branch(es) until we know if it'll work or not. Then what about position learning (PL)? Can it be utilized effectively with book learning? When we cut branches in case of failure, the chances of exploiting subsequent bad positions are slim. The alternative is to keep the PL files and reset the book learning after a while or after creating a new book. It make sense in the more gradual approach to learning, because we may travel the same line more than once. But I think the value is minimal at best. I think I'm more in favor of expanding the book through PL, ie. adding moves to the book according to succesful results. Another related option to the last suggestion is letting the engine create its own book through gameplay. I'm not interested in rejuvinating the tiresome debate about opening books or not. But I find the idea of creating a diverse opening tree through book learning interesting. Regards, Mogens
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.