Author: Dennis Breuker
Date: 02:54:11 01/20/04
Go up one level in this thread
On January 18, 2004 at 06:05:06, Tord Romstad wrote: >I am considering to implement PN search or one of the many related algorithms in >my >engine, in order to use it as a sort of "oracle" for the main search and eval. >The idea >is to formulate some sort of goal, and call the PN search to determine whether >this >goal could be achieved or not. For instance, in an endgame position a static >examination >of the position could show that white would win if he could manage to bring his >king to >g6 without allowing black to sacrifice his knight for white's h pawn, and the >main search >could ask the PN search whether this is possible. > >My hope is that ideas like this could be used to implement some sort of crude >long-range >planning in the endgame. > >Has anybody else experimented with similar ideas? Is it possible to make it >work? I haven't worked on something similar, but have worked with PN search and PN^2 search (see my thesis). I have thought about using PN search for finding material wins. In general, I think it is very difficult to use plain PN search for finding an arbitrary goal. PN search is guided by mobility: it prefers positions where the opponent has few moves, and you have many moves. That is why it is good in finding forced mates. If the opponent has few moves, there is a higher probability that there is a mate somewhere. And that is why PN search is not so good in finding mates where you have to make a "silent move" (don't know the Engelish term for it) somewhere in the sequence. A forced mate like 1.Nf7+ Kg8 2.Nh6++ Kh8 3.Qg8+ Rxg8 4.Nf7 mate is very easy to find for PN search, whereas a mate like 1.Bf6 any 2.Qh6 any 3.Qg7 mate is much more difficult to find for PN search. This means that when you have other goals than mate, you must guide the search for a most proving node by something else than mobility... And that's not easy. Dennis > >Tord
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.