Author: Uri Blass
Date: 12:31:02 10/03/03
Go up one level in this thread
On October 03, 2003 at 15:15:55, Anthony Cozzie wrote: >On October 03, 2003 at 14:43:24, Uri Blass wrote: > >>On October 03, 2003 at 13:27:20, Anthony Cozzie wrote: >> >>>On October 03, 2003 at 12:47:23, Uri Blass wrote: >>> >>>>On October 03, 2003 at 12:13:13, Anthony Cozzie wrote: >>>> >>>>>On October 03, 2003 at 12:03:49, Uri Blass wrote: >>>>> >>>>>> >>>>>>I can detect every big threat by adding a special search after making null move. >>>>>> >>>>>>For example if I like to detect threats of at least 2.2 pawns I can do >>>>>>val=-alphabeta(depth-4,-beta+219,-beta+220,...) >>>>>> >>>>>>if (val<=beta-220) >>>>>>threatmove[ply-1]=1; >>>>>>//ply-1 because I still did not undo the null move. >>>>> >>>>>small error here: beta should be replaced by eval, or you will have massive >>>>>instability problems. >>>> >>>>If I am a queen down in the search and I threat to win a bishop then I do not >>>>consider it as a threat because a threat is a threat relative to beta. >>>> >>>>I believe that Tord does the same(he replied that your example of threating the >>>>queen twice by sacrificing material is solved by extending only big threats so >>>>if you sacrifice a rook and a bishop the threat on the queen is not a big >>>>threat). >>>> >>>>I think that the idea that an extension should not be dependent on beta is a bad >>>>idea. >>>>It is better to have stronger engine with stability problems and not >>>>weaker engine without them. >>>> >>>>I prefer even not to care about using hash tables for pruning because my >>>>experience told me that I cannot get significant gain there easily(I have a lot >>>>of stuff that means that pruning or extension is not defined only by the >>>>position). >>>> >>>>Uri >>> >>>in your code: >>> >>>Suppose the BM extension succeeds, and the node fails low. In other words, the >>>program was using delaying moves to push a threat beyond the horizon, the BM >>>extension stopped it, and it realized it was in trouble (a reasonable scenario). >>> >>>Then beta will be (say) 200 less than it was previously, and the threat might >>>not be 200 less than beta any more, and your new re-search w/out the extension >>>will fail high. etc. >>> >>>Being a queen down in the search (may) have very little to do with beta: example >>>WAC#141, where beta is a mate score even though white is down a rook and a >>>queen, and beta for black is -mate score even though black is up a rook and a >>>queen. >>> >>>anthony >> >>I can agree with you that using beta may be a mistake because of the fact that >>beta is changed after fail high but using the static evaluation is also >>a bad idea. >> >>In the case of Wac141 I want to use instead of beta the last calculated score of >>the position that is almost equality that has nothing to do with the fact that I >>lost a queen and a rook. >> >>If I replace beta by eval I do not get it because I understand eval as the >>evaluation of the position that I have(I evaluate every node) >> >>Uri > >What I am saying is very simple: what is a threat? >Answer: a threat is the possibility, if the side to move passes, for the other >side to significantly improve their position. > >In other words, do a null search with eval-200, and fails low, it means that if >white passes black can get a position at least 2 pawns better than the current >position. > >anthony The question is not definitions. The point is that if I sacrifice a queen I may want to extend threat of mate but not threat of winning a bishop. I want to extend threat to get a position that is better than the position that I already know that I have based on 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.