Author: Robert Hyatt
Date: 20:35:33 06/05/00
Go up one level in this thread
On June 05, 2000 at 22:33:41, Bruce Moreland wrote: >On June 05, 2000 at 18:59:41, blass uri wrote: > >>On June 05, 2000 at 18:35:10, blass uri wrote: >> >><snipped> >>>[D]5k2/7K/6P1/1p6/1P3p2/1P6/8/8 b - - 0 1 >>> >>>In this position the static evaluation of Crafty is advantage for black because >>>it does not know that the white pawn cannot be stopped. >>> >>>It plays f3 with 4.96 pawns advantage for black at depth 1. >>>It is possible to know by evaluation that the white pawn also cannot be stopped >>>because the white king controls g7 and g8 >> >>It is possible to use the following simple rule: >>White pawn in the 6th rank is unstoppable if the white king control the queening >>square. >> >>It is possible to see also by evaluation that black will have to waste tempo >>and cannot promote the passed pawn because of the square of the black king. >> >>Uri > >This is a fine rule, but white doesn't get to this position since after Bxh6 >Kxh6, black can block the pawn if he's willing to give up his own pawn. > >I think it's important to given bonuses in positions like the one shown, where >the K is supporting the P, and not just in K+P endings, either. > >But really that's not the hard part of this. What ends up happening is that >black will figure out early on that he has to give up the f-pawn. That >resulting position is harder than this one. > >bruce I think there are two points that are causing some problems here... 1. a program will play the right move but not seeing the right "evaluation". This can be helped by a bonus for supported passers (I have such a bonus so that if the king is supporting an advanced passer it is more valuable by far than a passer out all by itself.) 2. a program will get the right evaluation. I worry more about getting (1) right _first_. IE the evaluation terms urge me in the right direction quickly, and then let the search find the real score. I worry about 2 later. Because you do want to recognize that you are winning or losing as soon as possible so that you can either advance toward the position or take evasive action if necessary. These kinds of positions are very hard to get right by 'special casing' the eval. I have a few "big terms" in my eval (two connected passers on 6th for example) but I try to wean the exceptions by requiring that neither be blockaded, and that the king not be close enough to help. Doing that will catch a bunch of damaging cases, but will miss a few. I think this sort of position falls into the same category. Tempo is very difficult to evaluate. This is a classic example where moving almost anything changes the result, including just changing the side to move. I am adding more and more special-cases it does seem, however. But I try to at least make them as general-purpose as possible.
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.