Asymetric king safety
Tom King
I've hear a lot about programs using "asymmetric" king safety in their
evaluation. Presumably, this is done to make the computer more paranoid
about king safety than the opponent - so the computer will play solid
defensive moves in preference to wild, attacking moves.
Is this a good deal?
I currently *don't* use an asymetric king safety function.
Comments?
--
Tom King
Robert Hyatt
: Hi all,
: Comments?
: --
: Tom King
I did it for a long time, the problem being that if you castle on the
same side with the opponent, and he plays Bxf6... you have a choice of
playing Bxf6 or gxf6. And gxf6 shreds your kingside pawn structure,
but it also opens a file on your opponent's king, which might be even
better for you.
Asymmetric scoring generally means that after calculating king safety for
both sides, you multiply the program's value by some value > 1, so that
its king safety is more important, and it will choose to play Bxf6 rather
than gxf6.
But then you get a 'passive' program, which is not bad. But not real
interesting. I've spent a lot of time trying to carefully analyze the
pawns, the open files and the enemy pieces to really understand which
king is more unsafe...
It has a way to go, but if you want 'interesting' you don't want
'asymmetric'. The only asymmetric thing I do now is blocked pawn
analysis to keep things 'open'.
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Tom King
<hy...@crafty.cis.uab.edu> writes
>Tom King <t...@hatbulb.demon.co.uk> wrote:
>: Hi all,
>
>: I've hear a lot about programs using "asymmetric" king safety in their
>: evaluation. Presumably, this is done to make the computer more paranoid
>: about king safety than the opponent - so the computer will play solid
>: defensive moves in preference to wild, attacking moves.
>
>: Is this a good deal?
>
>: I currently *don't* use an asymetric king safety function.
>
>: Comments?
>: --
>: Tom King
>
>I did it for a long time, the problem being that if you castle on the
>same side with the opponent, and he plays Bxf6... you have a choice of
>playing Bxf6 or gxf6. And gxf6 shreds your kingside pawn structure,
>but it also opens a file on your opponent's king, which might be even
>better for you.
This is very interesting. In these kind of positions, what do strong
humans play? Probably it depends on the exact position, who the player
is, whether they need a win or draw in the game, whether they are tired
or not etc. etc.
>
>Asymmetric scoring generally means that after calculating king safety for
>both sides, you multiply the program's value by some value > 1, so that
>its king safety is more important, and it will choose to play Bxf6 rather
>than gxf6.
>
>But then you get a 'passive' program, which is not bad. But not real
>interesting. I've spent a lot of time trying to carefully analyze the
>pawns, the open files and the enemy pieces to really understand which
>king is more unsafe...
>
Interesting. I really don't want to fall asleep watching my program
play, so for now, I'll stay symmetric.
>It has a way to go, but if you want 'interesting' you don't want
>'asymmetric'. The only asymmetric thing I do now is blocked pawn
>analysis to keep things 'open'.
>
Unless of course, by asymmetric you value the *opponent's* king safety
more the computer's. I can imagine that this might be:
- bad against very strong players, who won't drop material, will strip
the computer's king bare, then mate him
- great against all other human players. Why? Because the computer will
rip the opponent's king position to bits, and mate or win material. If
this involves weakening the computer's king position, so what? The
computer has sufficient tactical strength to fend off attacks from all
but the best players.
- "interesting" against other computer opponents. Can't decide this
one..
I wonder if CSTAL is more concerned about its safety, compared to the
opponent's? I'd guess CSTAL is uses asymmetrical king safety, where it
values the opponent's king safety more than its own. Hence the wild,
sacrificial attacks etc. I'm only guessing here..
--
Tom King
Steve Maughan
Richard Lang's Genius program is probably the biggest user of asymetric
evaluation and search (ie more selective with opponents moves than its own).
I would therefore agree that it will result in a solid/passive program much
like Genius.
Steve Maughan
Tom King
<maughanD...@bigfoot.com> writes
>>But then you get a 'passive' program, which is not bad.
>
>Richard Lang's Genius program is probably the biggest user of asymetric
>evaluation and search (ie more selective with opponents moves than its own).
>
Really?? More selective with opponent's moves than its own?? I don't own
Genius, but I'd have thought it was better to be selective the other way
round, i.e. always consider all opponent's moves, but be selective on
computer's moves.
>I would therefore agree that it will result in a solid/passive program much
>like Genius.
>
>Steve Maughan
>
>
>
>
--
Tom King
Steve Maughan
>>
>>Richard Lang's Genius program is probably the biggest user of asymetric
>>evaluation and search (ie more selective with opponents moves than its
own).
>>
>
>Really?? More selective with opponent's moves than its own?? I don't own
>Genius, but I'd have thought it was better to be selective the other way
>round, i.e. always consider all opponent's moves, but be selective on
>computer's moves.
Yep - meant to say more selective with own moves!
Steve Maughan