Author: Gerd Isenberg
Date: 08:51:42 11/21/03
Go up one level in this thread
On November 21, 2003 at 10:40:06, martin fierz wrote: >On November 21, 2003 at 08:26:56, Gerd Isenberg wrote: > >>On November 21, 2003 at 07:24:34, martin fierz wrote: >> >>>On November 20, 2003 at 15:20:59, Gerd Isenberg wrote: >>> >>>>Three Positions, one is mate in 13 (Shredder - IsiChess WMCCC02). >>>>Any static eval or king safety heuristic to choose the best one for white if >>>>leaf nodes ;-) ? >>> >>>i can tell you what i think when i get to see such a position in a game: this is >>>one where you can't rely on your feelings (=static evaluation), but have to >>>calculate! so basically i think the answer to your question is that no static >>>eval can catch this one. >> >>Hi Martin, >> >>yes, this is absolutely true in the sense to decide whether this is a won, drawn >>or lost position. Anyway, double defending controls on check trajectories is >>probably one heuristic fact to consider. >> >>>not very helpful i know... if anything, my gut feeling likes the position with >>>the pawn on d7 best, because you still have some kind of king shelter. that >>>would probably be a sensible rule if you are looking for one: keeping a pawn on >>>the seventh is good for your king. >>>as for the difference between position 2 and 3, that's just what we >>>german-speaking guys call "stellungsglück". >> >>Yes, but in (3) there is a safe c2-c8 trajectory break via Qc4, protected by d5. >> >>Cheers, >>Gerd > >so if i understand you properly you want to use a heuristic which >1) looks what checks are available and 2) goes on to see whether the opponent >can block that check with a piece of smaller or equal value than the checking >piece? sounds very expensive! besides it sounds quite inaccurate to me, because >perhaps (i) you can safely walk out of that check or (ii) you lose anyway when >you block the check (e.g. the piece that you need for blocking was defending >something else). such things should be resolved by the search IMO. Yes Martin, you are right of course, just some brain strorming about this kind of tiny heuristics and the fun to implement it with bitboards ;-) I do already a lot of expensive things in my king evaluation, including static mate in one (some in two or more) detection. I already have a "give check" target set for sliding pieces and getting the check trajetory (the squares from one target to the king) is only one lookup to a 64*64 bitboard table. Looking whether one trajetory has the "ward off" feature is only "and" with appropriate defend sets. I will see... Cheers, Gerd > >i certainly don't think in such terms at the chess board. those are the lines i >(try to...) calculate. the "pawn on 7th" for the king shelter is good if you >don't have the time to calculate as a human, and therefore probably good for the >static eval. > >cheers > martin
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.