Author: Robert Hyatt
Date: 13:42:03 01/24/99
Go up one level in this thread
On January 24, 1999 at 00:54:47, William Bryant wrote: >Thanks in advance. > >These specific questions have not been discussed recently her, or else I >missed the questions and responses. > >1. At the root (ply = initialply), you should not probe the hash table at the >start of the search. (source: crafty SearchRoot). > this is intuitive. You probe _before_ searching anything. If you get a 'hit' with EXACT score set, what move do you play? If you get a LOWER bound flag (fail high) which move fails high? etc... the info is useless at the root, so I just avoid doing it. Doesn't do a thing for time since this is only done 30-40 times per game.. >Is it OK to start probing the table at the first ply beyond the root? yes, because you now have a move played at the root.. and you can play that if the EXACT score comes up at ply=2 > >Although not probing the table at the root, do you back up you score and best >move when exiting the root search? I would think so, because this is where >draft will be the deepest as you exit the search. I'm not sure I follow that. What you store is the 'initial' depth on entry to Search, because that is the depth you will be at the next time you call search and do a probe. and you want these to match... > >2. I don't yet have an opening book, and start searching on move one. The >first implementation of my hash table began to significantly reduce search >time when the iteration reached 7 and 8, but the nodes per second didn't >change. I am assuming this is due to the fact that the hash table is causing a >cutoff and decreasing the number of nodes searched. Note: my search is to a >fixed depth at present. > >Is this expected behavior? >Should I see an increase in the NPS? >(in the opening phase of the game)? NPS won't change a lot.. but the 'quality' of the search does, because the deeper you search, the more likely it is you will get a transposition and 'hit' an older position. > >3. Any suggestions for common pitfalls in the implementation of a hash table >that >I should seek out. I think I've done this right. >The score and move are no different from my fixed search without the hash table, >but the PV moves at deeper plies are in a different order than from before. > >Any comment or suggestions would also be appreciated. if the score is identical, you should be ok. hashing can change the move ordering obviously, so the PV might be somewhat different, so long as the score doesn't change. And it turns out that the score can change... IE search fine 70 to 18 plies and you will get +1, because you can't win a pawn in 18 plies in that position. Search it with hashing to 18 plies and you might get +4, because you can possibly see deeper than you are searching because of the way the hash table grafts subtrees from one part of the tree to another... > >William >wbryant@ix.netcom.com
This page took 0.01 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.