Author: Robert Hyatt
Date: 16:26:04 07/09/02
Go up one level in this thread
On July 09, 2002 at 18:18:10, Uri Blass wrote: >On July 09, 2002 at 18:06:39, Robert Hyatt wrote: > >>On July 09, 2002 at 17:44:32, Uri Blass wrote: >> >>>On July 09, 2002 at 17:37:47, Robert Hyatt wrote: >>> >>>>>I agree that you can define null move pruning as extensions of the moves that >>>>>are not pruned but this is not what they did. >>>>> >>>>>Uri >>>> >>>> >>>>How can you make that statement? They extended the moves they thought >>>>important to extend. Null-move trims the depth on the moves it considers >>>>to be bad. >>>> >>>>They are the _same_... The effect is exactly the same... >>> >>>The moves that they extended were not the moves that null move search suggest >>>that they may be good so the effect is not the same. >>> >>>Uri >> >>So? You suggest that null-move is correct. I don't believe that at all. >>Neither did Hsu. Null-move makes mistakes. We just accept them because we >>like the overall effect. It is _certainly_ possible to extend the right moves >>and have fewer errors than null move which cuts off moves that appear to be >>bad when they are not, or allows them to be searched when they should not be. > >Null move without zunzwang verification does mistakes but their search did more >mistakes. > >I agree that it is possible to do something better but they did not do it. >They did not decide to extend moves if the move has a threat. Uri, you need to read some of the deep blue papers. They _did_ have a threat extension. Among several others. They did singular extensions. They did a shallower extension if there were two good moves rather than one (a special case of singular). They did a shallower extension if there were three good replies and the rest were all bad... The various papers they wrote explained a lot of this, including the one referenced at an online journal site in one of the DB threads here in the last week or two. > >They had different extensions rules. They certainly did, no argument there. But in comparing them to me, for example, they extended _far_ more things that I currently do. I believe that doing what they do at my hardware speed might be bad. But at 200M nodes per second, it didn't kill their depth enough and it let them see a lot of things I easily miss. > >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.