Author: Rafael Andrist
Date: 09:12:48 06/01/02
Go up one level in this thread
On June 01, 2002 at 11:35:31, Dan Andersson wrote: >The idea is similar to enumerating the state space in simple endgames. You will >select a won or unevaluated state for the attacker and a drawn or unevaluated >state for the defender. This will become an exact graph of the state space after >the iterations produce no more states. There are a few problems AFAIK. When will >you trigger the start of the algorithm? When is it cost effective, size of the >graph? How will you limit the enumeration, escape conditions? And it will work >much better when you have one attacker and one defender. In an ordinary game >search the position might be ambigious as to the eventual outcome and attacker >and defender might switch place due to a latent mate threat for example. The >idea will work wonders on fortress positions with locked pawns. But for more >dynamic positions with distinct conversions, a proof-number search or lambda >search will be more efficient. A PN^2 search will find a draw as easily as a win >given the right termination conditions. I doubt that a PN search will help because the king usualy has many possibilities to escape the checks (otherwise the perpetual check may be detected also by hash relatively fast). So the tree may explode and PN not be very helpful. Rafael B. Andrist
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.