Author: Uri Blass
Date: 05:20:03 07/20/02
Go up one level in this thread
On July 20, 2002 at 07:46:09, Sune Fischer wrote: >On July 20, 2002 at 06:59:50, Uri Blass wrote: > >>On July 20, 2002 at 06:43:44, Sune Fischer wrote: >> >>>On July 20, 2002 at 02:39:45, Uri Blass wrote: >>>>It is possible that they could get 2 productive changes: >>>> >>>>1)Add null move pruning >>>>2)Not use the singular extensions. >>>> >>>>It is possible that 2 is not productive without 1 and >>>>if they did not start by testing 2, they got the wrong conclusions. >>>> >>>>Did somebody try to test crafty with their algorithm >>>>(no null move pruning and singular extensions)? >>>> >>>>I guess that it is going to be clearly weaker than >>>>the Crafty of today at 120/40 time control. >>>> >>>>Uri >>> >>>I think IBM was playing it safe. >>>They could have taken more chances, adding nullmove probably would have made it >>>even stronger, but imagine the embarrasment for them if it failed to see a >>>simple combination like a mate in 1 :) >> >>I do not see how you can miss a mate in 1 with null move. >>You can miss mate in 2 but not mate in 1. > >All right, bad example, the zugzwang is better :) > >>It is also possible to do a verification search to reduced depth >>only to see that you do not miss a big zunzwang. > >Sure, but given the complexity of their project, perhaps they just tried to keep >some things simple. > >>The problem here is that changing the alpha and the beta value >>can give you wrong information in the history tables and >>in the killer moves so I may need to do a special search to >>reduced depth without updating the killer move and the >>history tables. >> >>Uri > >You've lost me there, the history table I use only for move ordering, and the >killers, well I try them if they are valid moves, but alpha-beta values? > >-S. My idea about zunzwang detection was to use different value of beta in the search that verify zunzwang because I care only about big zunzwangs and not about zunzwangs of 0.1 pawns. I found that it does movei slower for some reason and it is not the nodes that it searches to verify zunzwangs that are less than 1% of the nodes. I guess that the problem is some change in the tables that are updated by the search and these tables include killer moves and history tables. I do not know if this is the problem but I decided that if detecting zunzwangs without being slower by more than 1% is not trivial thing to do then I am not going to do it today. 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.