Author: Ratko V Tomic
Date: 00:00:50 09/20/00
Go up one level in this thread
> I also think that most of the positions will not be relevant in games. Depends on the game sample it is looking at. If you play frequently against a program, it would surely be helpful if the program would remember what it had found at, say, 30 sec/move level, earlier, instead of wasting users time to go over the same calculation (and routinely come up with the same answer) again in a previously seen non-book position. I often run into the same non-book positions against the program and would like program to remember what it decided at a certain level instead of wasting my time move after move. The so-called "book learning" doesn't quite work here since it is geared in the most simple-minded way to avoiding lines the program lost in before and repeating those in which it won. I said simple-minded since programs make no attempt to establish causal relation between the move they will avoid and the game result. You may win a game due to some far-away reasons, and program will avoid perfectly good post-opening line. Basically, the commercial chess programmers (or the marketing people who push for the features) are overly focused on comp-comp performance to care about such trivialities as wasting their customers time. Their scheme of "book-learning" and the "killer books" are prime indicators of where the chief priorities are. The best solution would be to have explicit setting for comp-human and comp-comp modes of play and adjust its behavior appropriately. Rebel 10 has some of this, but unfortunately not regarding the mode of position learning.
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.