Author: Uri Blass
Date: 00:43:00 03/07/02
Go up one level in this thread
On March 07, 2002 at 00:42:56, Slater Wold wrote: >On March 06, 2002 at 23:40:28, Robert Hyatt wrote: > >>On March 06, 2002 at 22:49:38, Slater Wold wrote: >> >>>On March 06, 2002 at 22:27:17, Uri Blass wrote: >>> >>>>On March 06, 2002 at 19:18:16, Slater Wold wrote: >>>> >>>>>Hyatt said in an earlier post that TB's don't take into account the ability to >>>>>castle because it would be a waste. >>>>> >>>>>However, when I feed this position into any engine, it solves it in 0.00 as a TB >>>>>win. >>>>> >>>>>[D]5Q1Q/5Q1Q/5Q1Q/5Q1Q/8/6P1/6k1/4KR1R w K - >>>> >>>>Your FEN is wrong and we need to imagine that all the white queens that you >>>>copied from dann corbit's post are missing. >>> >>> >>>You're right. But obviously this is not the position I am talking about, >>>because I don't have the 13 man TB's. :) >>> >>> >>>>>It shows 20 possible moves, all from TB's I am guessing. >>>>> >>>>>I cannot cut and paste the eval, because there isn't one, but I have: >>>><snipped> >>>>>1.+ - (#4) Rf4 >>>> >>>>This is not correct and the program that you use has bugs. >>>>It should not call tablebases in a position that is not in the >>>>tablebases(castling is legal) >>>> >>>> >_Several_ of these moves take castling into accout. >>>>> >>>>>After Rf4 Kxg3 my TB's show 28 moves. The first move is 1. + - (#2) O-O, the >>>>>last is 28. + - (#15) Rh8. >>>>> >>>>>I am 100% sure TB's do indeed take castling into consideration. >>>> >>>> >>>>No >>>>You do not understand how tablebases work. >>>>There are no moves in tablebases. >>>> >>>>The engine generates all the legal moves and looks in the tablebases after these >>>>moves to see distance to mate. >>> >>> >>>Um, well, according to Hyatt, it would tell the TB "o-o" and it wouldn't return >>>anything. I am very well aware how TB's work. >> >>Not quite. First I _never_ said anything like that. With EGTB's you don't >>give them a move, and get back a score, you give them a _position_ and you get >>back a score. And the score says "mate in N from the given position, >>assuming castling is impossible." >> >> >> >>> >>> >>>>If castling is legal then the engine looks at the tablebases to see the distance >>>>to mate after castling in order to see the mate in 2 score. >>> >>> >>>According to Hyatt, no it doesn't. >> >> >>You are greatly twisting things around. Re-read what Uri wrote... >> >>"if castling is legal then the engine only checks the TB _after_ castling >>has been done." Because after castling has been done, it can't be done again >>and the resulting EGTB score will be correct. Prior to castling, the score >>will be wrong because castling is possible but the EGTB scores don't include >>castling. >> >> >> >> >>> >>> >>>>Uri >>> >>>The correct position is: >>> >>>[D]8/8/8/8/8/6P1/6k1/4KR1R w K - >>> >>>Sorry. > >I am not trying to "twist" anything around. You said TB's don't take into >account castling. Ok, fine. I believe you. I open a chess engine, and it's >returning o-o as a TB move. NO EVAL NEEDED. I am asking a simple question. >Why?! How is an engine returning a mate, without TB's, without an eval? > >If it's a stupid question, I apologize. I just don't understand, obviously. Very simple answer 0-0 is not a tablebase move because the tablebases have no moves but only positions and evaluations. The position after 0-0 is a tablebase position. The engine tries every legal move and look at the tablebases to get a score for the position after the move. I can add that the way the engine is using tablebases is wrong and it can cause mistakes on positions when the only win is based on the idea to castle in the second move and not in the first move. In this case the engine may return draw score for every move because it is going to look at the tablebases that are based on the assumption that castling is illegal in the future. It is of course not important for practical games because castling is not allowed in practical games so I understand programmers who choose not to fix the problem. 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.