RE: Evaluation Methods for Programs

>===== Original Message From Noah Roberts <[EMAIL PROTECTED]> =====
>K wrote:
>> NR & Pham,
>>
>> Actually I have developed a Chinese Chess program for fun, based on the 
free
>> codes from Pham. At first I did it completely in Java, but later for speed 
I
>> change the engine to C and keep the GUI in Java (calling the C DLL via 
JNI).
>
>Perhapse you would be interested in my cxboard protocol:
>http://xiangqi-engine.sourceforge.net/cxboard.html
>
>>
>> My evaluation function is quite simple:
>> 1. material+position, i.e. a rook weight 154 originally, but 166 or more
>> somewhere,
>> 2. attacking score

      If you are attacking an enemy piece (irrespective of whether it is 
protected or not), you got an attacking score. In my program, I give score 
5,4,3,2,1 for attacking a rook,cannon,knight,elephant and bishop. Attacking 
king and pawn doesn't count.

>> 3. protecting score

      If you are protecting a friendly piece, you get a protecting score of 
1 
irrespective of the piece. I don't count pieces protected by king or rook 
(because rook should go out to attack).

>
>Can you explain these last 2 more?
>
>>
>> It can play reasonably good chess initially and midgame at 6-PLY even
>> without
>> opening book. However, the endgame is quite weak (cannot win with a rook
>> against a bare king and a passed pawn, it keeps chasing the pawn vertically
>> or
>> checking the king forever!!)
>
>This is a common problem in chess engines.  Endgame play requires some
>planning because the end is usually beyond the horizon.  Computers can't
>plan so by virtue they suck in the endgame.  That is why I would like to
>see a set of open source and standardized endgame tablebases.
>
>NR