Author: William Penn
Date: 09:23:08 03/29/05
Go up one level in this thread
On March 27, 2005 at 11:41:46, Steve Glanzfeld wrote: >Usually there is a certain evaluation difference required for position learning. >I don't know how Shredder 9 exactly does it, but I assume when it for example >analyses move A +0.20 for a while, and then finds the better move B +0.25, it >will most probably NOT learn. IIRC, I once saw a description of the position >learning function of another engine, which required evaluation jumps of at least >+/- 0.30 or bigger, to learn. So, I think that would remember a new move C +0.50 >(or better) in the example above, or when move A suddenly is evaluated -0.30 (or >worse) after deeper analysis. > >(0.30 is just an example value i seem to remember) > >Steve In my ongoing tests, and I've just had another positive confirmation today, it seems to be based purely on whether a capture is involved. If so, it remembers. If not, it remembers nothing. Sometimes there is only a small evaluation difference for different ways of capturing, and I haven't tried to determine if that is involved. But it is clear to me that it only works (remembers) when a capture is involved. Strange... WP
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.