Author: Ryan B.
Date: 17:38:23 02/14/06
Go up one level in this thread
On February 14, 2006 at 19:04:52, David B Weller wrote: >I was just thinking about knight forks, and thought of yet another way an engine >can 'delude itself' > >Lets say, white can fork black, but it is at the horizon. 0 depth remaining > >Now if black _HAS_ any captures, they will allow white to play its capture and >the gain of the fork will be realized > >But all black has to do, to make the position seem better, is retract any >threats it has, and poof! the fork disappears [search ends with white's knight >move, and it never gets to play the capture] > >I suspect there are a ton more rediculous things going on at the horizon that I >havent even thought of...what else is lost at the horizon!?!?! > >Delay type threats, capture/recaptures, hiding double attacks, .... > >I wonder, if there are so many different types of these things, and it isnt >really worth trying to catch the few I have found by, eg., recap extension, >threat extensions, etc... > >Thoughts? Good something related to computer chess. This is nice to see. One idea is to have a layered search. At horizon depth of your initial search have a no positional knowledge quick search for tactics. It should not be necessary to do this at the end of all lines, just PV + select number (possibly you can tune this with a Search Focus UCI parameter). Also make sure to hash if the node has been expanded this way. I think the idea is at least interesting. Ryan
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.