Author: Uri Blass
Date: 03:14:32 12/28/01
Go up one level in this thread
On December 28, 2001 at 05:36:52, Rafael Andrist wrote: >On December 27, 2001 at 18:36:33, Uri Blass wrote: > >>On December 27, 2001 at 16:51:52, Rafael Andrist wrote: >> >>>On December 27, 2001 at 15:47:33, Uri Blass wrote: >>> >>>>7 plies to stop the qsearch is not a holy number and >>>>I do not know the correct number but it is clear to me that >>>>not stopping the qsearch at some point is illogical and there is a position when >>>>Fritz needed an hour to find mate in 1 because of qsearch >>>>explosion. >>> >>>A possible solution which is theoretically correct but a bit tricky to >>>implement: stop qsearch after x plys and set an "incomplete" flag. If you >>>recognize during search that the "incomplete" node can still change the value of >>>the tree, do the qsearch again. (I don't do it yet.) >>> >>>> Limiting qsearch is, in my opinion, same as making >>>>>it almost useless. >>>> >>>> >>>>In most of the qsearches there are not lines of more >>>>than 7 plies so I do not see why limiting the qsearch to >>>>7 plies make it almost useless. >>> >>>if limiting the qsearch doesn't change much, why limit it and maybe lose >>>critical information? >> >> >>If in 100 cases it changes nothing and in 1 case I waste 100 more nodes because >>of not limiting the qsearch then the 1 case may be important. >> >>If the price that you pay to get the information is too high and the information >>is something that you relatively cannot trust then it may be better not to get >>the information. >> >>If I continue to give ideas for free then I have no chance to have a top program >>in the future so maybe it is better if I stop it. > >haha. I cannot see any new ideas from you in this thread. Limiting the qsearch >is also done by others. But it may be better for me to stop giving ideas for >free about theoretically sound ways to limit qsearch as above since you ignore >them. > >Rafael B. Andrist I understand the idea of the uncomplete flag but it is not clear to me that it is a good idea. If you set uncomplete flag only to repeat the search later then you may waste more nodes. I know that limiting the qsearch was done also by other people but the main idea that I read here was limiting the qsearch based on the search depth(only 2x plies in the qsearch when the nominal depth is x and I see no reason to decide about the limit based on the search depth) I believe that some commercial programs use similiar ideas (I guess that Genius and I guess that today also Tiger) but part of the commercial programs do not use it and the best free program with available source code(Crafty) does not limit the qsearch and I doubt if there is one free program that limit the qsearch based on the ideas that I suggested I mean: 1)Doing the search more selective in the last plies of the search for example by not considering pawn captures, 2)Having some limit to the sqearch based on the depth, 3)using special evaluation(SEE) when there are captures but you cannot search deeper because of your limit(SEE is used in crafty only for pruning captures in the qsearch or for order of moves and not for evaluating and this is the reason that I did not use the word SEE because I did not do a connection between SEE and evaluation function). Uri
This page took 0.02 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.