Author: Uri Blass
Date: 06:05:37 12/22/04
Go up one level in this thread
On December 22, 2004 at 08:18:21, Dieter Buerssner wrote: >On December 22, 2004 at 02:43:50, Harald Faber wrote: > >>The argument is easy and logical: >>The more positions you search, the more likely you reach positions from the TBs >>which are relevant for the current position. >>If you search for 3 sec per move it is unlikely that you hit the (right) >>tablebase positions than when thinking for 1 minute e.g. > >I don't think, this argument is correct. It is logical, that with only probing >at root, the engine must become better with TBs than without (at least when >assuming, that the oppent has TBs/ will play TB positions in the game >theoretical sense correctly). The default probing method of several engines seem >to hurt them. So one could conclude, they will probe too often. In identical >position, with a longer search, you would probe relatively more often. I suspect >somehow, that the probing shapes the search tree in some cases in some way, that >makes it less efficient, and that this is more decisive than the slow down >through the probing. I am aware, that this sound not really logical, and of >course I am aware, that a TB hit always cuts the tree. This sounds completely logical if you care to think about it. A case when not having tablebases is productive because a program may prefer losing KR vs KPPP and not drawn KR vs KPP was already mentioned. I suspect that the main problem is with draw scores and it is possible that it may be even better to continue to search in case of draw score and not to use tablebases later in that branch and to use tablebases only for mates in less than 50. 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.