Author: Jeremiah Penery
Date: 23:13:44 05/17/05
Go up one level in this thread
On May 17, 2005 at 01:53:19, Ricardo Gibert wrote: >On May 16, 2005 at 22:17:44, Jeremiah Penery wrote: > >>On May 16, 2005 at 16:18:37, Steven Edwards wrote: >> >>>On May 16, 2005 at 15:35:29, Matthew Hull wrote: >>>>On May 16, 2005 at 14:14:21, Steven Edwards wrote: >>> >>>>>The GA described so far is only for assisting with the suggesting of mating >>>>>attack moves. Other, more complex themes will require more complex pattern >>>>>matching including instance variables and nested patterns. >>>> >>>>I see. How is the current GA process progressing on the problem set? Also, >>>>does the Symbolic application run in a GUI, a console (or either)? >>> >>>The current GA run against BWTC has broduced a recognizer that picks the best >>>move in 722 of the 1,001 positions. This was produced after 320 breeding >>>cycles. >> >>Have you thrown out positions that are either wrong or have multiple 'best' >>moves? > >If you think about it, you will realize this doesn't matter much at all. As long >as the best move of "bad ones" represent a reasonable try. I agree that this is the case for a more general type of plan formulation, but not for a mate-solving plan. If there's no mate to be found, or if the keymove is simply wrong, you will be tuning based on bad information, which may be worse than tuning with no information. However, Steven's answer to the question was satisfactory, though I still have reservations about it, regardless of how small a percentage of the test problems really are wrong.
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.