Author: Angrim
Date: 10:17:16 06/19/01
Go up one level in this thread
On June 18, 2001 at 19:49:50, leonid wrote: >On June 18, 2001 at 15:59:36, Heiner Marxen wrote: > >>On June 18, 2001 at 03:45:49, Angrim wrote: >> >>>On June 17, 2001 at 16:55:53, Heiner Marxen wrote: >>> >>>>So let me ask: what is the status of your code? May we look at it? >>> >>>My code is private, sorry. However I would be glad to answer >>>questions about how pn^2 works. I have spent a fair amount of >>>time on pn^2 since it is very handy for checking the safety of >>>openning lines in suicide chess. I just had to plug in the >>>rules of chess to get a chess mate solver. >> >>Thanks for your offer! >>Next time when I try to understand or implement some sort of pn search, >>I will remember it and email you. >> >>Heiner > >If somebody between you will explain to me, in genral term, what it is all about >(pn search) will be very thankful. pn-search(proof number search) is a general purpose game tree proving heuristic. I will try to explain it here briefly, but I think that the thesis does a better job. Each node in a search tree is assigned a proof number and a disproof number, such that the proof number is the number of leaf nodes of the subtree which would need to be proven to prove the root of the subtree, and the disproof number is the number which would need to be disproven to disprove the root of the subtree. At each iteration in the search, it expands a leaf node of the search tree which would (if proven true) reduce the proof number of the root, and which (if proven false) would reduce the disproof number. If the leaf node is neither proven nor disproven, its proof and disproof numbers are calculated, and propagated back through the tree, possibly increasing the proof and/or disproof numbers at the root of the tree. When used for chess, a node is proven if the side to win delivers checkmate, and disproven if they are mated. Draws are a pain in the neck, but can generally be scored as having pn and dn of +infinity since a drawn position can never be proven to win for either side. Alternately a draw can be scored as a loss for the side that is trying to win, which makes things a bit simpler but makes the values asymetric. >Cheers, >Leonid. Angrim
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.