Author: Omid David Tabibi
Date: 23:03:05 05/11/03
Go up one level in this thread
On May 11, 2003 at 12:14:18, Uri Blass wrote: >On May 11, 2003 at 10:21:12, Omid David Tabibi wrote: > >>On May 11, 2003 at 07:19:04, Tim Foden wrote: >> >>>On May 10, 2003 at 21:23:00, Omid David Tabibi wrote: >>> >>>>On May 10, 2003 at 20:32:10, Matthias Gemuh wrote: >>>> >>>>>On May 10, 2003 at 17:49:01, Mike Siler wrote: >>>>> >>>>>>When Sjeng finishes a search, it displays among its stats a move ordering >>>>>>percentage. Does anyone know how this is calculate? >>>>>> >>>>>>Michael >>>>> >>>>> >>>>>In most programs, it is the ratio FailHigh_on_fist_move/All_FailHighs. >>>>> >>>> >>>>Why not use a more general figure: >>>>ratio of "first move being the best" / "all (interior) nodes" ? >>> >>>I think this is because there are generally 2 types of interior nodes... one's >>>that do fail-high, and ones that don't fail-high. >>> >>>In the ones that fail-high, we are very interested on the fail-high happening on >>>the first move. >>> >>>In the ones that don't fail-high, we generally fail-low (due to alpha+1=beta), >>>we don't (in general) have a best move. And it will make hardly any difference >>>what order we search the moves in, as we will have to look at them all anyway. >>> >> >>That's true, but my point is that we shouldn't confine the figure only to >>fail-high cases, but also consider other nodes which produce a best move. >>Anyway, the total figure will not vary significantly. > >I do not understand how do you know if a move is the best during a normal >search. > >Some facts: >1)If the first move fail high it does not mean that it is the best move and the >program only knows about a move that is good enough to refute the opponent move >and does not search for the best move that does it. Well, it means that the move ordering was also "good enough" :) > >2)If all the moves fail low you cannot know which one of them was the best >without special search. > >Uri
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.