Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old October 1st 07, 12:42 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Jul 2004
Posts: 834
Default Zap beats Rybka in close match..




help bot wrote:

I think there may be positions where Rybka holds her pawn
back, but should let loose, but by and large this aspect of
the program seems to help her in the endgame. In fact, her
obvious quirk of automatically placing pawns on the same
color as the opponent's Bishop seems to be a weakness,
yet in every game I have yet seen, this worked quite well.


I wonder whether placing pawns on the opposite color as
the opponent's Bishop in those situations would work better
or worse for Rybka? Could it be that she is compensating
for a weakness?

It seems that same color is a Good Thing while there are
many pieces on the board that benefit from a shield, and
that opposite color is a Good Thing when there are few
pieces on the board that could hinder the bishop from
attacking pawns from the rear. The question of exactly
where to switch from opposite to same is a very interesting
one. I know it when I see it, but how to write a program
that knows when to switch? Once the factors for making
that decision are decided on the weighting might be one
of those cases where a genetic algorithm gives a better
answer than a human can.

--
Guy Macon
http://www.guymacon.com/

  #2   Report Post  
Old October 1st 07, 03:57 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: May 2006
Posts: 9,302
Default Zap beats Rybka in close match..

On Sep 30, 6:42 pm, Guy Macon http://www.guymacon.com/ wrote:
help bot wrote:
I think there may be positions where Rybka holds her pawn
back, but should let loose, but by and large this aspect of
the program seems to help her in the endgame. In fact, her
obvious quirk of automatically placing pawns on the same
color as the opponent's Bishop seems to be a weakness,
yet in every game I have yet seen, this worked quite well.


I wonder whether placing pawns on the opposite color as
the opponent's Bishop in those situations would work better
or worse for Rybka? Could it be that she is compensating
for a weakness?



As far as I could see, Rybka was not compensating nor
calculating, but seemed instead to have been ordered by
the boss (i.e. programmer) to do this, regardless of any
subtle considerations like "is it more crucial /here/ to
restrict the opponent's mobility, or to make it such that
my own pawns are immune from attack?".

In every game I saw, the pawns were *automatically*
placed on the same color as the enemy Bishop, and
later the program did amazing things using zugzwangs
and what might pass for patience, determination,
resourcefulness, etc. These zugzwangs in particular,
seemed related to the restriction of the enemy, yet it
is hard to imagine that with a King, a Rook and a Bishop
the enemy was so restricted as to be forced to lose. I
think this is similar to those games where Rybka slowly
maneuvered out of perpetual check, or finessed her own
King gradually toward the enemy's, leading to a winning
attack (except against Zap Chess, which usually drew)
where the King (most often) assisted her Queen in a
direct assault on the enemy King.


It seems that same color is a Good Thing while there are
many pieces on the board that benefit from a shield, and
that opposite color is a Good Thing when there are few
pieces on the board that could hinder the bishop from
attacking pawns from the rear. The question of exactly
where to switch from opposite to same is a very interesting
one. I know it when I see it, but how to write a program
that knows when to switch?


As far as I could see, Rybka solved this dilemma by
automatically preferring restriction over pawn-safety.


Once the factors for making
that decision are decided on the weighting might be one
of those cases where a genetic algorithm gives a better
answer than a human can.


Another such item is that Rybka does seem to prefer
Bishops over Knights, but this comes into play in the
opening, whereas the other issue primarily affects the
endgame phase.

I find the suggestion that a "genetic algorithm" can
give a better solution than a human can interesting; on
another forum, GM Suba commented that in some
endings, "all computers" play like beginners. My view
is that these most recent games may well have been
affected by a fast time control, rather than both Rybka
and Zappa playing like "beginners" because they have
horrible endgame algorithms. Given sufficient time,
the programs seem to play the endgame very well in
all but a few cases, and even there we may not realize
the implications of an operator-selected contempt
factor, which I would blame on the operator. Also note
that on occasion, they talk about one program having
table-bases while the other had none (or had less). As
a human, I can say with absolute certainty that we all
play like "beginners" when compared to this standard.


-- help bot

  #3   Report Post  
Old October 1st 07, 07:03 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Jul 2004
Posts: 834
Default Zap beats Rybka in close match..




help bot wrote:

I find the suggestion that a "genetic algorithm" can
give a better solution than a human can interesting;


It has been proven to be so in certain, limited non-chess
applications:

"For example, a genetic algorithm developed jointly by
engineers from General Electric and Rensselaer Polytechnic
Institute produced a high-performance jet engine turbine
design that was three times better than a human-designed
configuration and 50% better than a configuration
designed by an expert system by successfully navigating
a solution space containing more than 10^387 possibilities.
Conventional methods for designing such turbines are a
central part of engineering projects that can take up
to five years and cost over $2 billion; the genetic
algorithm discovered this solution after two days on
a typical engineering desktop workstation." (Ref below)


The big limitation is that it works best to optimize a
value (such as when to switch from blocking a bishop
with your pawn to putting your pawns on the other color
where they are safer) but pretty much useless for a
situation like what you described -- blindly always
keeping the pawns on the same color as the opposing
bishop -- or in the case where the program doesn't
factor in pawn/bishop color. It takes a human to
figure out that same color is good erly and opposite
color is good late, but once the human does that, a
genetic alorithm is likely to be better at figuring
out the exact place to swirch. There are many other
places where a GA might do well, such as deciding
when to switch from king hiding to king attacking.

I haven't been able to track it down, but I seem to
remember someone programming a Backgammon game with
genetic algorithms determining the strength of various
decision factors. As I recall, it got stronger and
stronger as it played against slightly different
variations of itself, eventually getting to the point
that it played better than the best they were able to
do with a human tweaking the factors and test-playing.


References:
http://www.cs.cmu.edu/Groups/AI/html...netic/top.html
http://www.talkorigins.org/faqs/gena...html#strengths
http://www.talkorigins.org/faqs/gena...ml#limitations
http://en.wikipedia.org/wiki/Genetic_algorithm

--
Guy Macon
http://www.guymacon.com/


  #4   Report Post  
Old October 1st 07, 07:59 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: May 2006
Posts: 9,302
Default Zap beats Rybka in close match..

On Oct 1, 1:03 am, Guy Macon http://www.guymacon.com/ wrote:
help bot wrote:
I find the suggestion that a "genetic algorithm" can
give a better solution than a human can interesting;


It has been proven to be so in certain, limited non-chess
applications:

"For example, a genetic algorithm developed jointly by
engineers from General Electric and Rensselaer Polytechnic
Institute produced a high-performance jet engine turbine
design that was three times better than a human-designed
configuration and 50% better than a configuration
designed by an expert system by successfully navigating
a solution space containing more than 10^387 possibilities.



When I was a very young bot, they had gasoline-
powered lawn mowers which were very loud, dangerous,
inefficient, unreliable and hard to start, averaging around
3.5 horsepower.
Today, we have gasoline-powered lawn mowers which
are very loud, dangerous, inefficient, unreliable and hard to
start, but progress has been made in that average
horsepower is up to, let's say 5 horsepower. (The folks at
GE and RP need to get in touch with Briggs and Stratton.)


In chess, the brute force method seems to have won
out; look at endgame table-bases, for instance, or a
description of the top programs, which will invariably
say: "brute force with alpha-beta pruning". The losers
cop out by suggesting the competition is cheating by
tuning programs to beat other top programs, and even
when the number one program loses, a plethora of
excuses are presented ("we used an old version", or
"a new bug was suddenly discovered").

Although I would not go so far as to class the top
programs as "beginners" in some positions, I do think
we still have a very long way to go before approaching
perfection in chess, and one possibility is that a
program or algorithm could one day tweak a chess
program far better than any human could. (Look at
what happened when humans attempted to tweak
Rybka for the match against GM Benjamin: disaster!)



-- help bot






Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Zap beats Rybka in close match.. SAT W-7 rec.games.chess.misc (Chess General) 43 October 2nd 07 01:41 AM
Zap beats Rybka in close match.. Guy Macon rec.games.chess.computer (Computer Chess) 7 October 1st 07 03:23 AM
Zap beats Rybka in close match.. Guy Macon rec.games.chess.computer (Computer Chess) 7 October 1st 07 03:21 AM
2nd Annual RGCP Grudge Match Sam Sloan alt.chess (Alternative Chess Group) 29 April 26th 06 07:50 AM
FIRST ANNUAL RGCP GRUDGE MATCH IS TODAY - SUNDAY, JUNE 26, 2005 (details inside) RGCP Grudge Match alt.chess (Alternative Chess Group) 1 June 26th 05 05:49 AM


All times are GMT +1. The time now is 11:08 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017