Author: Robert Hyatt
Date: 12:40:41 11/09/00
Go up one level in this thread
On November 09, 2000 at 07:35:19, Joe Besogn wrote: > >Go ahead. I'm never surprised by anything, so show me those special cases you >didn't code up. KR vs KPP. First, let's get on the same page. I am _not_ talking about KR vs KPP. I am talking about positions where one side has two pawns on the 6th. The other side might have a rook, might have a knight, might have a bishop + rook, who knows. I am simply speculating in my evaluation, that when I reach the 6th with two connected passers. I have a significant advantage. As other criteria are met, the advantage goes up (ie neither pawn is blockaded, bonus++, if the king is outside the square of the pawns so that it can't help at all, bonus++ again. Etc. But the bonus is there even with queens on the board, just not as big. It is still there with RB for the opponent. And there can be pawns elsewhere that are influencing the game in ways it doesn't yet understand. So not KPP vs KR. but K*PP vs K*. I might have sacrificed a rook to get the two connected passers (ie in wac 2 crafty solves it at ply=3). I might have sacrificed a knight. I might have only sacrificed a single pawn. Any of those can be dangerous sacrifices if the pawns do get stopped. > >>I've been burned by them. Do I have to show you the exceptions >>where the bishop trapped on a2/etc does _not_ win? > >I never said it did. This is a risky case. You have to guess at it. Better is to >code for all trapped situations, not the patzer trap on a2. > >My king safety can range >>from -3 to +3 roughly. Should I show you cases where it is wrong? > >Yes please. Show me a nice complicated position you score +3 for king safety and >it is wrong. Or maybe even -3? That is more common. Because I am asymmetrical. > >> What >>about those cases where an outside passed pawn _doesn't_ win? What about >>those cases where a distant majority doesn't win? >> >>Lots of risk, lots of speculation. Of course the 'degree' of speculation can >>be controlled if specific exceptions are recognized. But if you look at the >>pawns on the 6th, note that this does _not_ require that the king be out of >>the square. To get the full +5 score, yes. But not to get a score that will >>tempt it to sac a piece for a pawn or two to reach such a position. I simply >>watch for exceptions and try to code to handle them as they are found. And >>slowly the speculative terms become more and more accurate. But I certainly >>have _lots_ of speculation that needs more control. > >Pawn 'speculation' or not-speculation is too easy. There can't be an effective >program around that doesn't play with scores for passed pawns and connected >passed pawns. The precise bonus values are indeterminate, they affect program >style of play more than anything else. I hope you don't try fiddling the values >with streams of test positions trying to get it accurate to two places of >decimals? I don't worry about decimel places, to be sure. But I do look at test cases when I add something new. I try my best to visualize the examples where I want it to work, and then try to dream up similar positions where it ought not work to be sure that I catch them. And then I watch the GMs on ICC uncover the special cases that I missed. Which always happens.
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.