Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Ply depths?

Author: Angrim

Date: 18:28:52 03/08/02

Go up one level in this thread


On March 08, 2002 at 01:18:25, Uri Blass wrote:

>On March 07, 2002 at 16:44:42, Angrim wrote:
>
>>On March 06, 2002 at 12:00:02, Miguel A. Ballicora wrote:
>>>Interesting. My program Gaviota finds Mate in 15, with the same sequence
>>>of moves in 18 seconds on a PIV 1.4 Ghz. Also, at ply 6 and I am very happy with
>>>that. The interesting part is that Gaviota requires ~2Mega positions
>>>where CM requires only 100K. In this particular position CM's tree is
>>>20-fold leaner! I will have to work hard on trimming the fat!
>>
>>Or, you could just add a proof-number search stage.  This type of
>>position is very easy for proof-number search.
>>
>>pn-search on athlon 1.2ghz:
>>fen: 7k/2p3p1/4Q2p/p3n3/2pBP2P/2P3P1/1r2q3/5RK1 w - - 0 1
>>proved that move f1f8 wins, 14.5 turns
>>PN:17176 evals, 1260 expands,  0.06 seconds
>>
>>Perhaps the most interesting part of this is that my pn-searcher has
>>no chess-specific knowledge other than how to generate legal moves
>>and score a mate position.  I expect that someone who is good at
>>chess could write a version that is much better by writing an eval that
>>measures how close to being proven each leaf node is.  I just use
>>the game-independant measure of counting the legal moves at each
>>leaf node.
>
>
>Can you explain what is proof number search and what is the difference between
>it and singular extension.
>
>This position is easy to solve with singular extensions because in every move
>black has a single reply to prevent mate(and in most of the cases there is only
>one legal move or only one legal move to prevent mate in 1 when the only
>exception is 3...Kxf6 when black has the alternative 3...Kh5).
>
>I also see that you evaluate 17176 nodes.
>I am interested to see the full tree.
>
>What is your order of moves?
>
>Do you start with moves that reduce the number of legal moves of the opponent.
>
>Does it mean that the first moves to search for white are 1.Rf8+ and 1.Qg8+
>By this logic after 1.Rf8+ Kh7 you have 3 candidates(2.Qg8+,2.Qg6+,2.Rh8+)
>After 2.Qg8+ Kg6 you have Qf7+ first because Rf6+ gives black 2 replies.
>
>
>
>Uri

See http://www.cs.vu.nl/~victor/thesis.html
also do a search on google for "proof number search"

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.