Author: Amir Ban
Date: 02:13:32 04/23/98
Go up one level in this thread
On April 23, 1998 at 02:57:52, blass uri wrote: > >I am sorry I did not explain myself clearly. > >my idea is: >1)when Junior4.6 analyses a move that it "thinks" is not best >the depth will be the same. > >2)when Junior4.6 analyses a move it "thinks" is best then the depth >will be the sum of one ply and depth d-1(usually something like d+1) > >3)if Junior discover a new move is best the depth will be the same as in >2 > >from my experience in correspondence there are cases that this method >is good (of course it can do some demage but I think if we use this only >for big d's it usually helps). > >a similiar idea can be used in other programs but we should define >another depth d-.5 at them when at depth d-.5 the computer do what >it usually do at depth d-1 except after the best move when it computes >one ply +depth d-1. Ok. Translated to programmers terms this reads: At the root, moves can be rejected by search to regular length, but to accept a move, you must extend and recalculate it. This is a valid idea, whose purpose is to get a longer and more reliable PV. There are probably people in this newsgroup who tried this at some stage, so they can comment. My own view is that current search algorithms anyway are optimized to reject moves fast, and spend most of their time on the move that will be played. This suggestion will skew the time distribution even more. Besides, candidate moves will now have to pass a double test to be accepted, so they odds are against them. I'm guessing that this will prove to be counter-productive, unless sonmeone can show otherwise. Amir
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.