Author: Ulrich Tuerke
Date: 08:51:04 03/03/03
Go up one level in this thread
On March 03, 2003 at 11:21:02, Uri Blass wrote: >On March 03, 2003 at 10:52:54, Ulrich Tuerke wrote: > >>On March 03, 2003 at 10:09:46, Albert Bertilsson wrote: >> >>>On March 03, 2003 at 09:43:24, Robert Hyatt wrote: >>> >>>>On March 03, 2003 at 09:13:36, Albert Bertilsson wrote: >>>> >>>>>Hi! >>>>> >>>>>During a recent game my engine reported that a mate in 2 was found, then after >>>>>the next opponent move it reported that a mate in 3 was found. Does this >>>>>indicate a bug in the engine? >>>>> >>>>>If it matters the opponent didn't followed the PV that engine displayed mate in >>>>>2. But after the mate in 3 was claimed the opponent followed the PV until its >>>>>defeat. >>>>> >>>>>/Regards Albert >>>> >>>> >>>>It could be >>>> >>>>(a) a bug in hashing, where you don't adjust the mate score before storing it, >>>>or you >>>>don't adjust it when you hit it in the table. >>>> >>>>(b) your extensions make it easier to find the mate in 3 than the mate in 2 and >>>>you quit >>>>too soon. >>>I interpret your answer as this should not happen if the engine has no bugs and >>>no extensions. Since Sharper doesn't have any extensions it would be (a) then... >>>alhtough I can't see how this could happen. >> >>I think that this can happen without extensions, too. >> >>It's a consequence of alpha-beta. Often some move of ply n+1 already suffices to >>refute the preceeding move of ply n, instead of the checking move leading to the >>mate. It's rather a matter of move ordering. >> >>It may for instance help a bit in some cases to prefer checking moves when >>ordering moves. >> >>Uli > >I do not see how it is possible. > >The first mate was mate in 2 and it means that in the next position there is a >mate in 1. > >A program should never miss mate in 1 and prefer mate in 3. >The move that delivers the mate is supposed to cause an immediate cutoff. > >Uri I confused things. The score of the best move is reliable. A consesquence of AB is that you don't know whether the remaining moves (2nd best and so on) had been refuted because they were only slightly worse or even mates. So, a 9 ply search should always be able to find a mate in 4 - neglecting extensions and pruning issues). Uli
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.