Author: Uri Blass
Date: 21:53:54 06/04/03
Go up one level in this thread
On June 05, 2003 at 00:04:37, Michel Langeveld wrote:
>On June 04, 2003 at 19:50:06, Anthony Cozzie wrote:
>
>>On June 04, 2003 at 16:30:45, Michel Langeveld wrote:
>>
>>>>>//Nullmover with only material
>>>>>Nodes per sec.: 319485
>>>>
>>>>>//Nullmover with everything switched on:
>>>>>Nodes per sec.: 301770
>>>>
>>>>This looks strange. I think you should expect a much bigger difference in
>>>>nodes/s.
>>>>
>>>>Regards,
>>>>Dieter
>>>
>>>Yes I was also thinking it was odd...
>>>I expected also a bigger difference ... Am not sure why this is ...
>>>
>>>my eval starts with:
>>>
>>>int generatePositionScore(int alpha, int beta)
>>>{
>>> evalCounter++;
>>>
>>> if (isDraw()) return 0;
>>> }
>>>
>>> scoreType score = p.materialWhite - p.materialBlack;
>>>
>>> //lazy eval
>>> if (score - 150 > beta) return score;
>>> if (score + 150 < alpha) return score;
>>>
>>> ...
>>>}
>>>
>>>It can be that the first part which I call lazy eval catches a lot of positions
>>>.... or there is a much bigger hotspot as the evaluation currently.
>>>I will run a profile to see if I can find more out ....
>>
>>Kingsafety in many programs can be over 1.5 pawns.
>>
>>Almost any reasonable evaluation of passed pawns will require corrections of
>>over 1.5 pawns.
>
>Yes, you are right. Kingsafety and passed pawns is still on the todolist of
>Nullmover. It simply doesn't have any kind of such code. I have to take care to
>rise these scores after implementing it.
I think that even without passed pawn evaluation you can get more than 1.5 pawns
evaluation only by piece square tables.
If one side has 2 pawns in the 7th,Rook in the 7th and knights and bishop in the
centre when the other side has knights in the corners you may get it.
It may not be common enough to justify not using lazy evaluation
but if you have mobility scores and scores against isolated pawns then you can
get easily more than a positional score of 1.5 pawns by the sum of positional
factors.
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.