Author: Thomas Mayer
Date: 11:18:52 02/01/05
Go up one level in this thread
Hi Eiko, >With up to 5 (or 6) men you can always test your code against the tablebases >(just query all possible positions ;) ). right, but then I must program something like that... :) But I might do in the future... because it is very interesting... >I've been trying to make some rule-based "tablebases" some time ago, but even >KBN-K seemed quite difficult... well, I see this as a combination of search and bitbase oriented knowledge... At the moment I am quite enthusiastic, I believe that the final code will solve studies no other program can solve yet, with TBs or without... I had already some amazing results with the knowledge of 3men bitbases and some rules... (Which might not all be correct at the moment) Slowly it seems also to have an impact on Quarks endgame strength... Just an example (which is of course not strength relevant) what we can do with 4men tablebases: [D] 5k2/8/7P/7P/7P/2KB3P/7P/8 b - - 0 1 Well, we all see at first sight that this is draw, programs have there problems (but of course would usually all chose the correct move) Now we have only KBPK-bitbase available... Well, I check now the bitbases with the most advanced pawn, if that draws, the hole position is draw. You can also do the opposite: KBP*K (* star means always multiple, in this case multiple pawns) Of course this is usually won (with some programmed exceptions) - but how to be sure that it is won, we do not have the correct bitbase ? Simple: Check the bitbase with each pawn, if you find a win, the hole position is won... So I differentiate here between maybe won and for sure won, so I can score the position different. E.g. our research may say that the position should be won, but we haven't found yet a win -> so I return a score that is not very high and search deeper. As soon as I have found a win, I can score way higher and cutoff. If I can find no win I will return a quite low score -> and that way I am maybe quite nearer to the truth... The search will help here and I narrow the tree a lot... When I have programmed my ideas I may present some results of this work... Greets, Thomas
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.