Author: Vincent Diepeveen
Date: 11:20:19 11/29/99
Go up one level in this thread
On November 29, 1999 at 12:06:57, William Bryant wrote: >A thought occured to me (which can be a dangerous thing). > >When a null move fails high (produces a cutoff), this value is stored in the >hash table. Should the depth of the hash table entry be the current depth >of the search, or the depth used for the Null search (ie depth - null >reduction). In contradiction to Fail High reductions you can store it in the hashtable without problems. Fail High reductions return the score of the same search with a smaller depth. However nullmove returns the score of a smaller depth where THE OTHER SIDE has the move, so that's something quite different, as the other side gets the move twice! The last few plies nullmove is a pure threat detection (depending upon how well your quiescencesearch is), however the plies before that it's also positionally seeing everything as it does a normal search. The room for error is that it is using a reduced search depth. However considering a certain side may move twice there is already a lot less room for error as when someone is threatening something, making 2 moves in a row is *definitely* speeding things up a lot. The pseudo code which is completely correct: int alfabeta(int alfa,int beta,int depth,...) { if( depth <= 0 ) return(qsearch(alfa,beta)); .. NULLMOVE: nullmovescore = -alfabeta(-beta,-beta+1,depth-1-R,..); if( nullmovescore >= beta ) { StoreHash(alfa,beta,nullmovescore,depth); return(nullmovescore); } .. .. } the only thing we can fight about is what the qsearch should see and what it should not see. The better the qsearch the more you see the last few plies. Suppose you don't do checks in qsearch, then you won't detect a matethreat last 1+R plies! How important is it to see a matethreat the last few plies? Everyone must answer that for himselve. >I have traditionally used depth, and am now testing with a depth - null depth to >see if it helps/hurts. > >I would appreciate comments from others who have tried or considere this. > >William >wbryant@ix.netcom.com
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.