Geschrieben von:/Posted by: Sune Fischer at 22 March 2004 18:23:18:
Als Antwort auf:/In reply to: Re: Important announcement Re:Waster geschrieben von:/posted by: Dann Corbit at 22 March 2004 17:31:48:
Yes. We won't let you off the hook so easily, Geoff! When you have released
your engine to the public, there's no easy way out. As long as there is still
any way to reach you, we will keep pestering you with requests for new and
improved versions.
Just replace the evaluation function with something you write yourself. Initially
it will perhaps be weaker than the current version of Waster, but quite soon you
will have something stronger than ever.
Here is an easy way to mathematically calculate piece square tables for pieces:
For each square on the board, see how many squares the piece would attack if the board were empty and the piece sat on that square. You will instantly see why a knight is bad in the corner.
I recommend keeping the bonus for proximity to the enemy camp outside of the piece square tables. For instance, the value of knight placement will depend on what side the opponent has castled.
You can do the same thing for pawns but it will only show you that the two edge pawns are terrible weaklings.
You must make a bonus for the two center pawns. I think that they block the bishops from controlling and also support other pieces that you want to place in the center. You should try several numbers (perhaps with a binary search).
These piece square tables will be better than the ones that you find laying around.
Trick:
For the kings, invert the piece square table in the opening and use it normally in the endgame. That is because ability to attack is also congruent to ability to be attacked. In the beginning, you don't want your king attacking anything unless that is the only choice.
This doesn't work so well in my program.
I think it is because rooks and bishops are better as long range attackers.
They are able to support and attack the center while still being safe at home
in the bunker.
In later stages of the game it probably changes, rooks and queens can more
safely go to the center without being driven around by pawns and minors.
A binary search?
I think the problem is that it requires lots of testing for each setting.
Knowning that these changes are bound to be minor it's also evident it's going
to be quite hard to see the difference.
It's not so easy to make good piece squares, sometimes one is surprised at
what should be working doesn't work.
I think piece squares are too static to be used in a king safety eval. Sometimes it is suicide to castle into the corner.
Probably better to try and evaluate if there is a good
pawn shelter and not too much attacking power on it.
This general safety code should be enough to encourage the king to castle
to safety, IMO.
-S.
