Author: Christophe Theron
Date: 13:01:04 01/01/98
Go up one level in this thread
On January 01, 1998 at 00:18:08, Don Dailey wrote:
>But let's consider this possibility too: Asymetrical evaluation!
>It's not hard to imagine that a programmer who concieves of the
>asymetrical search style of Genius might also do similar things
>with evaluation. King safety: weight heavier if computer king
>in trouble, passed pawns: opponents more dangerous than computers,
>Bad pawns: more serious if belonging to computer, etc. In fact
>I think this could be tested with some carefully constructed tests.
Yes, I'm quite sure that Genius evaluate advanced passed pawns
differently for it than for you. Just make the following experiment: set
on the board a white pawn on 7th rank, the white king very close, the
black king quite far, and a black rook on 8th. Set ply depth 1 and
selective search 0. This will force Genius to compute 2 plies lines.
Make it play both sides and watch carefully the lines and the score. At
a critical moment you will see Genius evaluate the line to about -2 for
him playing black, and the next move to about 0 for him playing white.
If it doesn't work, try setting a slightly different position. Sooner or
later, you'll get a position were it works. I have done the experiment,
but didn't save the exact position, sorry.
This way you can convince yourself that it has an asymetrical evaluation
for those pawns. Thorsten, I'm sure you are going to try it yourself!
This test has nothing to do with asymetrical search. The search is
probably asymetrical, but in a way that is still not clear to me. I
don't know if Thorsten has the right answer, but I'm still convinced
that such a search could affect the playing style. It's obvious (for me)
that if you prune heavily on your side you get a less agressive play.
You can even prove from a mathematical point of view that the minimax
score you get with this search is always less than or equal to the
standard minimax score.
So there are positions where standard minimax gets a huge score (a
combination is found), and the "prune heavy for me" search doesn't find
the combination. That's all Thorsten wanted to say. It's very simple,
even for him that never wrote a chess program.
You can also find positions where a standard minimax finds a slight
positional advantage (maybe an agressive move), and the asymetrical
search doesn't.
That's what we want to say by "affect the playing style". Nothing more.
Of course changing the evaluation is a much easier way to control the
playing style. But remember that Lang's programs had a hard life in the
early days, trying to compete against the very fast Spracklen's
programs. The passive style is surely a side effect of the asymetrical
search that was designed to resist the strong tactical opponents.
>Also, doesn't Genius have a full width mode now in some versions?
>Turn this off and do some positional assymetry tests. Whatever
>the answer is about positional assymetry, I'll bet Genius will not
>suddenly become an aggressive attacking player in full width mode.
The point is to find out if in the normal playing style Genius does
asymetry in the first plies or not. I suppose Thorsten can find
positions where this is obvious. Give him some time to publish such
positions.
Christophe
This page took 0.01 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.