Author: Miguel A. Ballicora
Date: 12:10:59 05/19/01
Go up one level in this thread
On May 18, 2001 at 23:21:07, Robert Hyatt wrote: >On May 18, 2001 at 15:54:51, Dan Andersson wrote: > >>I believe we are arguing somewhat the same thing here. If you have a dynamic >>measure (not based on forward search, but search history) with a predictive >>power of 90% (whatever) accuracy, would you not use it to decide when to enable >>a costly extension (that might fail). >> >>Regards Dan Andersson > > >No... because the idea of singular extensions is to extend those moves where >there is only one good choice and everything else is significantly worse. I >don't see how some history-based "trigger" would work very well here. IE you >have to test a move for singularity in order to know whether or not it is >singular. I'm not quite sure what I might try to use to determine that the >test is irrelevant, other than the few obvious ideas already known... ie no >point in doing a SE test if you are in check and you only have one legal move >to get out. No point in doing a SE if you are recapturing a piece to restore >the material balance. No point in doing a SE if the singular move is a piece >attacked by the opponent at the previous ply (otherwise the opponent could blow >the search up any time it is advantagous)... > >IE I can think of history ideas on when to not do it... but that leaves a _lot_ >of open ground where they will be tried... trying to restrict it more strongly >by trying to predict which moves _will_ be singular seems hard... > >IE I read the two choices as (a) do like DB and test most all moves for >singularity; (b) do something different and test a very few moves for >singularity. (b) seems harder and more dangerous... I am a bit confused and I have some questions. First, it is not clear to me one thing about what you say: For you, (b) is incorrect or you do not feel it is a good idea? That is important to me because if it is the latter it leaves more room to me for experimentation. Why do you say it is dangerous? specifically what do you have in mind that can go wrong? What I have in mind it is not only to detect statically when there is a good chance that there will be a singular move, but also 1) to detect what kind of position will be more affected for the presence of a singular move 2) positions that the test would be cheaper to do (i.e. less moves available like a response to a check) Questions related to SE in general: Is it tried close to the leaves? because in the leaves I cannot reduce by 2 the search Is it used on top of other extensions? like recapture etc.? Regards, Miguel
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.