Author: Robert Hyatt
Date: 20:28:50 01/30/02
Go up one level in this thread
On January 30, 2002 at 11:52:45, Albert Silver wrote: >On January 29, 2002 at 23:53:06, Robert Hyatt wrote: > >>On January 29, 2002 at 06:40:12, Albert Silver wrote: >> >>>On January 29, 2002 at 04:31:39, Jeroen van Dorp wrote: >>> >>>>Convecta claims to ave them (complete) on 9 CD's. Are there still some missing >>>>or so? >>>> >>>> >>>>http://store.convekta.com/shop_model.asp?gid=121&sView=Catalog >>>> >>>>J. >>> >>>A few piece combinations were indeed removed as they took up space for no good >>>reason. The combinations in question are King + 3 queens against King, King + 2 >>>queens + rook against King, etc. They represent no practical value to either the >>>user or the engine and removing them saves from 1-2 Gb of HD space. >>> >> >> >>They do have _some_ merit. If the engine probes after a capture takes it to >>5 or less pieces, then you want to get a hit. If you omit some tables, some >>of the probes will fail, wasting time. And causing extra search. > >I don't understand. If the search at that point doesn't get a tablebase hit with >a KQQQ vs. K and instead gets a score of +30 pawns, it will somehow contrive to >play worse?? Or make a poorer choice? Say opt for an alternate line that gives >it +35 pawns instead of the direct mate? > > Albert First, that screws up the alpha/beta search. If _all_ the scores are near a mate score, and then a few drop way down, that widens the search window and makes the search go slower. If you are already winning, fine. But if you are not, this can hurt. Also you can look silly... would you _really_ want to hit a KNN vs KP ending and possibly draw, because the tables say MATE in N, rather than hitting a KQQQ vs K which says +30 rather than mate in 9 or whatever? > >> Yes a >>program should win KQQQK or KRRRK or KBBBK or KNNNK, but it helps to know >>_exactly_ how deep the mate is, to make the search as efficient as possible.
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.