Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 8 WAC toughies at long time control - WAC 230

Author: Robert Hyatt

Date: 12:21:18 03/16/01

Go up one level in this thread


On March 16, 2001 at 13:51:43, Dann Corbit wrote:

>On March 16, 2001 at 09:47:09, Robert Hyatt wrote:
>[snip]
>>Blocked pawns are not so hard to find.  Crafty has some code that does this
>>extremely accurately, to try to prevent these from happening.  It builds a
>>big bitmap of squares that pawns can move to, which takes into account how
>>many pawns attack and defend each square.  This way it doesn't have to have
>>two pawns rammed together to know they are blocked.
>>
>>It is still hard to figure out what to do, of course.  Right now I simply
>>penalize the score when there are lots of pawns and few or no lever
>>possibilities...
>
>Why not simply set the eval to 0.0 when you have a complete blockade, or eval -
>piece_value when you have just a blocked piece?  That way, the eval function
>will automatically find ways to unblock it.

It isn't so easy.  Sometimes blocking the position is the _right_ thing to
do, otherwise you end up losing rather than drawing.  That is the problem
I was thinking of above...




>
>Benefits:
>If you are behind, the engine will try to keep it blocked.  That's what you
>would want to happen in this situation.  So it helps you fight for the draw.  If
>you are ahead, then it will consider the value of crashing the barricade to be
>just slightly less than your current material advantage.  Again, I think this is
>exactly what it ought to do.

The problem is more complex than you think.  Your base score is +.4, so you
try to avoid blocking things up and by doing so you let the score start to
slip away.  But it is now too late to block (you missed your one chance).

These kinds of positions are very difficult to handle when a computer is really
not sure about whether the resulting positions are drawn (a good thing) or
drawn (a bad thing).




>
>So if there is a wall between you and your opponent, then the eval is 0.0.  I
>think that solves all the wall problems.



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.