Hiarcs6 full of bugs???
Harald
Another strange behaviour appeared, no matter if engine or DOS-version.
Take a look at this position:
8/8/2Kb4/1n6/8/8/8/4b1k1 w - -
Hiarcs realizes that Kxb5 is best.
Now change position and place the knight from b5 to b6 and look what
Hiarcs plays.
Isn't it strange behaviour that is hard to explain?
I don't know how far these things influence its play, maybe it would give
Hiarcs some more points to correct these bugs, maybe I am wrong.
Komputer Korner
Harald wrote:
I don't know what you are talking about. I set this position up in
Hiarcs 6 DOS and it played Kxd6.
The position is so weird anyway with 2 dark square bishops.
--
Best regards
Komputer Korner
The inkompetent komputer
If you see a 1 in my email address, take it out.
Note that my true email is still kor...@netcom.ca
I don't often check the email of the sympatico address.
Harald
quoting a mail from kor...@netcom.ca
KK> > Another strange behaviour appeared, no matter if engine or
KK> > DOS-version.
KK> > Take a look at this position:
KK> >
KK> > 8/8/2Kb4/1n6/8/8/8/4b1k1 w - -
KK> >
KK> > Hiarcs realizes that Kxb5 is best.
KK> > Now change position and place the knight from b5 to b6 and look what
KK> > Hiarcs plays.
KK> > Isn't it strange behaviour that is hard to explain?
KK> >
KK> > I don't know how far these things influence its play, maybe it would
KK> > give
KK> > Hiarcs some more points to correct these bugs, maybe I am wrong.
KK>
KK> I don't know what you are talking about. I set this position up in
KK> Hiarcs 6 DOS and it played Kxd6.
In both cases?? Nb5 and Nb6??
If this is so, you have a different version, I got confirmation that there
are no viruses on the 3 computers. Can anybody explain the different
results??
KK> The position is so weird anyway with 2 dark square bishops.
That's the point. :-)
Harald Faber
Komputer Korner
chrisw wrote:
> --
>
> >
>
> I've not looked at this position, but .........
>
> Hiarcs is a knowledge program, particularly in the endgame. My
> experience
> of chess knowledge programming is that every knowledge chunk you put
> in
> will have unintended side effects. Slowly, as the side-effects are
> recognised in play testing, the side-effects will get taken care of
> (and
> quite possibly other side effects get so created).
>
> The more knowledge chunks - the more the side effects.
> snipped
> My guess is that you'll find many strange behaviours in these program
> types. Just wait and see my program when it gets released :)
>
> Chris Whittington
>
> >
> > Harald Faber
> >
Having 2 dark squared bishops is so rare that I wouldn't worry about
whether the program realizes that they can't mate or not. If everybody
was in an AI race with chess programs, I would worry about it, but the
chess programmers are in a chess race not an AI race. That was given up
a long time ago.
chrisw
--
http://www.demon.co.uk/oxford-soft
Harald <dirty...@sensecom.de> wrote in article
<6a0cd...@29.sensecom.de>...
I've not looked at this position, but .........
Hiarcs is a knowledge program, particularly in the endgame. My experience
of chess knowledge programming is that every knowledge chunk you put in
will have unintended side effects. Slowly, as the side-effects are
recognised in play testing, the side-effects will get taken care of (and
quite possibly other side effects get so created).
The more knowledge chunks - the more the side effects.
So, you can have it either way: brute force materialism which creates no
anomolies but screws up if the search can't 'find' the positional knowledge
that's lacking; or knowledge chunks which normally deal with the
positional, but screw up sometimes. And since the coding of these programs
is on-going, the strange behaviour will always be manifesting itself, now
and into the future.
Komputer Korner
Harald wrote:
> qKK> The position is so weird anyway with 2 dark square bishops.
>
> That's the point. :-)
>
> Harald Faber
Even if this happens, this case is so rare that it is not worth
worrying about. Harald, you have to realize that there always be
positions even for Deeper Blue where it doesn't understand what is going
on. The tradeoff is that these positions are rare so that they don't
affect the long run score. Often correcting these anomalies will make
the program play weaker in the long run.
brucemo
Komputer Korner wrote:
> Having 2 dark squared bishops is so rare that I wouldn't worry about
> whether the program realizes that they can't mate or not. If everybody
> was in an AI race with chess programs, I would worry about it, but the
> chess programmers are in a chess race not an AI race. That was given up
> a long time ago.
This is quite correct. If the goal is to look smart in every case, or to
try to stump people who like to fool chess programs, or to try to avoid
losing that one game in a million, it would be fine to do this ending, but
otherwise it is simply not important. This one won't come up.
If you want to continue the discussion with a practical case, you can
chose the ending with bishop and wrong rook pawn versus nothing. That
comes up lots, and is really worth Elo points if you handle it right.
bruce
brucemo
chrisw wrote:
> I've not looked at this position, but .........
I did. If anyone else wants to avoid setting it up, the position involves a
choice between capturing a bishop and capturing a knight. If you capture the
bishop, you are in a lost KBN vs K ending. If you capture the knight, you
are in a drawn KBB v K ending, it's drawn because the bishops are on the same
color squares.
> Hiarcs is a knowledge program, particularly in the endgame. My experience
> of chess knowledge programming is that every knowledge chunk you put in
> will have unintended side effects. Slowly, as the side-effects are
> recognised in play testing, the side-effects will get taken care of (and
> quite possibly other side effects get so created).
Great example is bishop versus two pawns. If you have the pawns, what score
do you give this position? Is it -1 or plus something? It is hard to make a
program understand that the bishop can't win this, without also causing it to
do something like sacrifice its bishop in the strong side of a KBPP vs KBP,
in order to reach a KPP vs KB that might be less winnable than the ending
with the material still on. It seems incredibly stupid, any human can handle
this quite easily, but it's hard to know how to deal with this in a general
evaluation function.
> The more knowledge chunks - the more the side effects.
>
> So, you can have it either way: brute force materialism which creates no
> anomolies but screws up if the search can't 'find' the positional knowledge
> that's lacking; or knowledge chunks which normally deal with the
> positional, but screw up sometimes. And since the coding of these programs
> is on-going, the strange behaviour will always be manifesting itself, now
> and into the future.
The trick isn't writing the code to handle the dorky cases, that isn't what
slows you down, and a lot of times the code isn't hard to write, the trick is
knowing that you are in a dorky case. The code that detects these cases is
not very complicated, but if you write the wrong code and put it in the wrong
place, it can slow you down a lot.
I think that a lot of these weird endgame cases do come up. This specific
one, with the two bishops on the same color, doesn't come up, but there are
some, like the "impotent pair", that come up a lot. If you don't code for
these cases, you will blow a lot of half-points. It's amazing how often you
can sneak into a drawn pawnless ending, or a drawn pawn-down ending.
A lot of the strange cases can be solved with a general solution -- endgame
tables. But that's expensive from a disk-space (and sometimes search speed)
point of view, and it comes with its own set of problems.
bruce
Dan Thies
On Sat, 19 Jul 1997 00:03:42 -0700, brucemo <bru...@seanet.com>
wrote:
>chrisw wrote:
>> Hiarcs is a knowledge program, particularly in the endgame. My experience
>> of chess knowledge programming is that every knowledge chunk you put in
>> will have unintended side effects. Slowly, as the side-effects are
>> recognised in play testing, the side-effects will get taken care of (and
>> quite possibly other side effects get so created).
>
>Great example is bishop versus two pawns. If you have the pawns, what score
>do you give this position? Is it -1 or plus something? It is hard to make a
>program understand that the bishop can't win this, without also causing it to
>do something like sacrifice its bishop in the strong side of a KBPP vs KBP,
>in order to reach a KPP vs KB that might be less winnable than the ending
>with the material still on. It seems incredibly stupid, any human can handle
>this quite easily, but it's hard to know how to deal with this in a general
>evaluation function.
It's pretty easy with a rule-based approach to handle cases like this.
In a general evaluation function, it probably could be done, but the
main side effect is probably slower search.
You could probably get around some of the speed issues by building the
evaluation from scratch before every search, and not adding the code
for "dorky cases" unless there's a chance of needing it.
Dan
Komputer Korner
Dan Thies wrote:
>
>
> You could probably get around some of the speed issues by building the
>
> evaluation from scratch before every search, and not adding the code
> for "dorky cases" unless there's a chance of needing it.
>
> Dan
Wouldn't building the evaluation algorithm from scratch before every
search slow down the program even more? And if you meant evaluate at the
root like Nimzo does then you suffer the resultant search problems of a
root processor.
Harald
quoting a mail from kor...@netcom.ca
KK> > qKK> The position is so weird anyway with 2 dark square bishops.
KK> >
KK> > That's the point. :-)
KK> > Harald Faber
KK> Even if this happens, this case is so rare that it is not worth
KK> worrying about. Harald, you have to realize that there always be
KK> positions even for Deeper Blue where it doesn't understand what is going
KK> on. The tradeoff is that these positions are rare so that they don't
KK> affect the long run score. Often correcting these anomalies will make
KK> the program play weaker in the long run.
Hmm. I'm still wondering about the difference if Nb6 or Nb5. Why the one
is solved correctly when the other one is not.
More interesting is the other position where H6 doesn't realize this mate
in one.
Harald Faber
Komputer Korner
Harald wrote:
>
>
> Hmm. I'm still wondering about the difference if Nb6 or Nb5. Why the
> one
> is solved correctly when the other one is not.
> More interesting is the other position where H6 doesn't realize this
> mate
> in one.
>
> Harald Faber
The other position had only 1 legal move. Harald, search for some
meaningful positions.
Jonathan Berry
In article <33D066...@seanet.com>, brucemo <bru...@seanet.com> wrote:
:Great example is bishop versus two pawns. If you have the pawns, what score
:do you give this position? Is it -1 or plus something? It is hard to make a
:program understand that the bishop can't win this, without also causing it to
:do something like sacrifice its bishop in the strong side of a KBPP vs KBP,
:in order to reach a KPP vs KB that might be less winnable than the ending
:with the material still on. It seems incredibly stupid, any human can handle
:this quite easily, but it's hard to know how to deal with this in a general
:evaluation function.
At the risk of advocating a return to the Dark Ages--
What happens if your evaluation function produces a vector
rather than a scalar? In other words, the evaluation might
have factors for opponent's minimum score, maximum score,
expected score and then a breakdown of score by material,
positional values etc.
For KPP vs KB, the opponent's maximum score would be "draw"
and the minimum "loss" or "work it out" or "look it up in the
tablebase".
When your program arrived at KBPP vs KBP, evaluating a superior
position, it would not be interested in the "opponent's maximum
score as draw" factor of the evaluation of KPP vs KB. However,
in some other ending which it evaluated as inferior, it would
be very interested in that factor.
Another possible benefit: When a computer is ahead in material
but the opponent has the initiative, many programs (not as many
as used to be) will grab onto a second pawn or a second piece
to increase their total "advantage". A human in the same
situation would be inclined to spread his advantage among
several baskets. For example, if he is a piece ahead, he might
sacrifice the Exchange to regain the initiative. A program
that has a single number to work with will not so easily make
such a decision.
I don't think these factors would be worked out to 5 decimal
places. 8-bit integers would usually suffice.
A human (well, this human) looks at a typical situation where a
B/g5 can take a N/f6 and double the f-pawns (g7xf6). There's a
tradeoff: I double the pawns, he gets the bishop pair. All
kinds of factors come into play, but often a critical one is
the safety of Black's king. OK, so king safety is always
critical, but not quite so at the heart of the matter, for
example, in the case where you're considering Bb5xNc6. A
multi-factored evaluation might allow the, hmm, evaluation
manager, to zero in on specific factors that are of vital
consequence when considering an earlier transaction.
I guess nodes per second would take a huge hit.
--
cheers,
Jonathan
Komputer Korner
Jonathan Berry wrote:
> I don't think these factors would be worked out to 5 decimal
> places. 8-bit integers would usually suffice.
>
> A human (well, this human) looks at a typical situation where a
> B/g5 can take a N/f6 and double the f-pawns (g7xf6). There's a
> tradeoff: I double the pawns, he gets the bishop pair. All
> kinds of factors come into play, but often a critical one is
> the safety of Black's king. OK, so king safety is always
> critical, but not quite so at the heart of the matter, for
> example, in the case where you're considering Bb5xNc6. A
> multi-factored evaluation might allow the, hmm, evaluation
> manager, to zero in on specific factors that are of vital
> consequence when considering an earlier transaction.
>
> I guess nodes per second would take a huge hit.
>
> --
> cheers,
> Jonathan
The unreleased program CSTAL has the feature where the evaluation
score is broken up into 12 different values to get a final total
evaluation score. The beta I saw was somewhat asymmetric however.
chrisw
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<33D54A27...@netcom.ca>...
> Jonathan Berry wrote:
>
> > I don't think these factors would be worked out to 5 decimal
> > places. 8-bit integers would usually suffice.
> >
> > A human (well, this human) looks at a typical situation where a
> > B/g5 can take a N/f6 and double the f-pawns (g7xf6). There's a
> > tradeoff: I double the pawns, he gets the bishop pair. All
> > kinds of factors come into play, but often a critical one is
> > the safety of Black's king. OK, so king safety is always
> > critical, but not quite so at the heart of the matter, for
> > example, in the case where you're considering Bb5xNc6. A
> > multi-factored evaluation might allow the, hmm, evaluation
> > manager, to zero in on specific factors that are of vital
> > consequence when considering an earlier transaction.
> >
> > I guess nodes per second would take a huge hit.
> >
> > --
> > cheers,
> > Jonathan
>
> The unreleased program CSTAL has the feature where the evaluation
> score is broken up into 12 different values to get a final total
> evaluation score.
The evaluation of CSTal is made up from an uncounted number of components
(1000+, I'ld guess).
For your convenience, these are grouped into various chess-idea components
(12 values) and can be displayed on screen - for any move at the root, and
for all of the moves in the PV.
>The beta I saw was somewhat asymmetric however.
The evaluation function itself isn't asymmetric (apart from bugs), and
except for one or two cases. The search is more strongly asymmetric.
Chris Whittington
mclane
mclane:
What do you understand under the words "somewhat asymmetric however" ?
Komputer Korner
mclane wrote:
Please no, not the Asymmetric controversy again. I suffered enough
flames over it.
For a method of how to test for asymmetry, see the bottom of the text.
For programmers' views on asymmetry see below:
": My views...
: 1.. Don't look at *the score* as being the absolute truth!
: 2.. *The score* is just an indication, nothing more, nothing less...
: 3.. Scores are introduced by programmers to find the best move.
I agree. There is no reason to take evaluations like -0.19 too
literally.
When I use computer programs to analyse my games, I only look for
tactical
errors. Most of these are found quickly, the asymmetrical evaluation
does
not matter. Computers are not good tools for finding positional moves.
: About asymmetry...
: Many chess programmers use asymmetry which is smart if done properly
but
: there are many dangerous side effects which prevented me till now to
: implement asymmetry in Rebel.
: Possible useful asymmetry subjects...
: 1.. King attack (different bonus/pentalties for white/black)
: 2.. If a program thinks it is in a lost position you can take your
last chance
: and put EVERYTHING on King attack or on a certain passed pawn.
: Very human alike :)
: 3.. If a program thinks it is in a lost postion and suddenly discovers
a
: repetition give it a score of 0.50 and not the usual 0.00
: Maybe there is more than just a draw?
: 4.. If a program thinks it is is in a lost position than every
checking move gets a
: bonus in the search tree till a certain maximum. The idea: maybe there
: is a repetition after the horizon?
: 5.. As point 4, avoid checking moves in a won position by giving a
: certain penalty for every checking move of the opponent.
: 6..
: 99.. Avoid closed positions by adding a bonus for pxp if the board is
: still full with pawns...
: Ok, I have given 6 asymmetrical subjects.
: Who is next? :)
-Ok, here are two:
-7.. Give a penalty for piece exchanges when material is equal. Most
programs
play better in the middlegame than in the endgame, therefore avoiding
exchanges might be a good idea (espescially in play agains humans).
-8.. Give a bonus for "crazy" positions (positions with lots of pieces
hanging,
bad pawn structures for both players, both kings insecure etc.). The
point
is to steer towards tactical positions, which the program usually plays
better.
--I use both these ideas in my own program. Number 7 works very well
(probably
because my program is the worst program ever in the endgame), number 8
seems
to work against humans, but not against other programs. The reason is
simple:
Tactical positions favour (of course) the player who is strongest
tactically.
My program is tactically stronger than most humans, but much weaker than
the
commercial programs.
--The last two weeks, I have also made some experiments with
asymmetrical
search. I recently replaced the null-move algorithm with a
knowledge-based
selective search algorithm. This still doesn't work very well, the
program
makes too many tactial oversights (and I am afraid Ed doesn't want to
tell
me how to avoid this :-)). To minimize the effects of these oversights,
I
have tried using different search algorithms for odd/even depths. I have
experimented with the following approaches:
1. Using excactly the same pruning algorithm at all nodes.
2. Prune away moves only at odd ply depths, search all moves at even
ply depths.
3. Prune away moves only at even ply depths, search all moves at odd
ply depths.
4. Prune at all nodes, but prune away more moves at odd depths.
5. Opposite of 4.
-In fact, I now have four different search functions in my program:
search_odd(), search(even), quiesc_odd(), quiesc_even(), which call each
other using mutual recursion.
-I still don't know which method works best, but they give quite
different
playing styles!
-I would like to hear the opinions of other people using knowledge-based
pruning. Do you prune asymmetrically? Do you search selectively from ply
1, or do you search a few plies full-width first?"
KK's View
An asymmetric program deliberately skews the evaluation of a position
towards one side because of a basic weakness in it's ability to ward
off long range attacks or a weakness for running it's pawns on the
kingside if the opponent has done so. There are other situations as
detailed above but it is basically a method to make the program think
that it
is better off than it is so that the program will initiate some
attacking ideas or in the reverse, make the program think that it is
worse off than it is, so that it will defend better. In other words
asymmetry
makes the program play more exciting attacking chess or play better
defense
in other situations. . In fact asymmettry often helps the program to
play
better chess. See above for the cases where it helps. The downside is
that
those asymmetric programs are notoriously difficult to change without
affecting other playing aspects so that one tiny change can have many
unforseen consequences. Asymmetry isn't completely to blame here but it
does exacerbate the problem. The other major problem with asymmetry is
that no matter how good the computer's scoring function gets, if there
is asymmetric code affecting the evaluation, the user will never be able
to get a true picture of the evaluation of the position. That doesn't
matter so much at the present time, because scoring evaluators have up
to now been woefully inaccurate, but that could change as programs
approach
perfection in positional chess.
HOW TO TEST FOR ASYMMETRY
"Set up a position, and then let the program search for N plies, then
set
it up again, play the move the program chose and let it search from the
other side for N-1 plies, and compare the evaluations..."
brucemo
Komputer Korner wrote:
> The downside is
> that
> those asymmetric programs are notoriously difficult to change without
> affecting other playing aspects so that one tiny change can have many
> unforseen consequences. Asymmetry isn't completely to blame here but it
> does exacerbate the problem.
I have not found this to be true. If anything the reverse is true.
bruce
chrisw
--
http://www.demon.co.uk/oxford-soft
brucemo <bru...@seanet.com> wrote in article <33E207...@seanet.com>...
> Komputer Korner wrote:
> > The downside is
> > that
> > those asymmetric programs are notoriously difficult to change without
> > affecting other playing aspects so that one tiny change can have many
> > unforseen consequences. Asymmetry isn't completely to blame here but it
> > does exacerbate the problem.
>
> I have not found this to be true. If anything the reverse is true.
Indeed. KK talking bollocks again.
Can KK please inform us just how he knows that
"> > that"
"> > those asymmetric programs are notoriously difficult to change without"
"> > affecting other playing aspects so that one tiny change can have many"
"> > unforseen consequences. " ?
I mean, has he tried it himself, this change of an asymmetric program ?
Which person who has carried out such changes has informed him of of this
factoid ?
Which is this group of asymmetric programs that are so "notoriously
difficult to change" ?
And before you answer KK, let me give it you:
large programs tend to be difficult to change.
badly written programs tend to be difficult to change.
kludged, mod-ed and hacked programs tend to be difficult to change.
asymmetry has nothing to do with it.
Chris Whittington
>
> bruce
>
chrisw
--
http://www.demon.co.uk/oxford-soft
mclane <mcl...@prima.ruhr.de> wrote in article
<5rta5c$m2k$1...@steve.prima.ruhr.de>...
> Komputer Korner <kor...@netcom.ca> wrote:
>
> Sorry but I am not capable of reading your mixed statements, quotes
> and all this stuff.
>
>
>
> >HOW TO TEST FOR ASYMMETRY
> >"Set up a position, and then let the program search for N plies, then
> >set
> >it up again, play the move the program chose and let it search from the
> >other side for N-1 plies, and compare the evaluations..."
>
> Play a game of chess with program a and write down evaluations and
> main-lines.
> A has e.g. white.
> After the game:
> Setup the beginning-position but let black have the first move.
> Let program a
> now make the first black move. Try to replay the whole game with
> inverse colores.
>
> If the scores or the main-lines are much different ....
>
>
> You can use games or the bt2630i for invers.
>
> Still I dson't know why you call chris program asymmetric. The main
> problems with it occur because it is a)
> very selective
> b) evaluating speculative
> c) often evaluating DREAMS and CHANCES instead of
> real-accurate-main-lines.
>
> It is the job of the opponent to destroy these dreams.
I don't want my dreams destroyed :(
Chris Whittington
Komputer Korner
chrisw wrote:
>
>
> Which is this group of asymmetric programs that are so "notoriously
> difficult to change" ?
>
> And before you answer KK, let me give it you:
>
> large programs tend to be difficult to change.
>
> badly written programs tend to be difficult to change.
>
> kludged, mod-ed and hacked programs tend to be difficult to change.
>
> asymmetry has nothing to do with it.
>
> Chris Whittington
>
> >
> > bruce
> >
If you have an evaluation function that has x number of variables with
parameters, and you then add a global adjustment to the score of that
evaluation function, what you have done is overidden the individual
parameters and substituted another function on top of the previous one.
Now you are 2 levels removed from the variables themselves. So when you
make a change to one of the parameters on either level, how can you
isolate the rest of the parameters when the function on the other level
is affecting their score? Or is your asymmetric code done differently
than this?
mclane
Komputer Korner <kor...@netcom.ca> wrote:
Sorry but I am not capable of reading your mixed statements, quotes
and all this stuff.
>HOW TO TEST FOR ASYMMETRY
>"Set up a position, and then let the program search for N plies, then
>set
>it up again, play the move the program chose and let it search from the
>other side for N-1 plies, and compare the evaluations..."
Play a game of chess with program a and write down evaluations and
main-lines.
A has e.g. white.
After the game:
Setup the beginning-position but let black have the first move.
Let program a
now make the first black move. Try to replay the whole game with
inverse colores.
If the scores or the main-lines are much different ....
You can use games or the bt2630i for invers.
Still I dson't know why you call chris program asymmetric. The main
problems with it occur because it is a)
very selective
b) evaluating speculative
c) often evaluating DREAMS and CHANCES instead of
real-accurate-main-lines.
It is the job of the opponent to destroy these dreams.
>--
Komputer Korner
mclane wrote:
> Sorry but I am not capable of reading your mixed statements, quotes
> and all this stuff.
>
> >HOW TO TEST FOR ASYMMETRY
> >"Set up a position, and then let the program search for N plies, then
>
> >set
> >it up again, play the move the program chose and let it search from
> the
> >other side for N-1 plies, and compare the evaluations..."
>
> Play a game of chess with program a and write down evaluations and
> main-lines.
> A has e.g. white.
> After the game:
> Setup the beginning-position but let black have the first move.
> Let program a
> now make the first black move. Try to replay the whole game with
> inverse colores.
>
> If the scores or the main-lines are much different ....
>
> You can use games or the bt2630i for invers.
>
Chris's program is such an exception to the rule that this discussion
is moot. Don't worry about it.
BTW your test for asymmetry doesn't accomplish a thing. If CSTAL's
scores were different in your test, I would really start to worry for
it. Your 2nd game is only a mirror image of the first. The first London
tournament was played like that, but there is no difference in score
evaluation from mirror images.
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
: chrisw wrote:
: >
: >
: > Which is this group of asymmetric programs that are so "notoriously
: > difficult to change" ?
: >
: > And before you answer KK, let me give it you:
: >
: > large programs tend to be difficult to change.
: >
: > badly written programs tend to be difficult to change.
: >
: > kludged, mod-ed and hacked programs tend to be difficult to change.
: >
: > asymmetry has nothing to do with it.
: >
: > Chris Whittington
: >
: > >
: > > bruce
: > >
: If you have an evaluation function that has x number of variables with
: parameters, and you then add a global adjustment to the score of that
: evaluation function, what you have done is overidden the individual
: parameters and substituted another function on top of the previous one.
: Now you are 2 levels removed from the variables themselves. So when you
: make a change to one of the parameters on either level, how can you
: isolate the rest of the parameters when the function on the other level
: is affecting their score? Or is your asymmetric code done differently
: than this?
In Crafty, all scoring is initially processed in a fully symmetric manner.
Then, at the end, Crafty says "Hey, I'm playing white, so my king safety
is more important than my opponents". This is a simple multiply done at
the end of things. Up to this point, everything is the same for both
sides... I've even played with a symmetric Crafty, and all that was needed
was to remove the asymmetric multiplier. So symmetric or asymmetric doesn't
necessarily make anything simpler nor more complicated...
Dave
KK, you're right about knowledge-based programs, in the past. They've
been hard to fine-tune.
At present, we have crafty programmers, with a Tal like genius for
CHESS, and ferreting out these problems and shredding them. One rebel
yell from programmers like these and your hair will fritz up, but the
problem will be deeply thought through, and a chess challenger these
modifications no longer need be.
Happy Friday, everyone!
Dave
Komputer Korner
Robert Hyatt wrote:
>
>
> In Crafty, all scoring is initially processed in a fully symmetric
> manner.
> Then, at the end, Crafty says "Hey, I'm playing white, so my king
> safety
> is more important than my opponents". This is a simple multiply done
> at
> the end of things. Up to this point, everything is the same for both
> sides... I've even played with a symmetric Crafty, and all that was
> needed
> was to remove the asymmetric multiplier. So symmetric or asymmetric
> doesn't
> necessarily make anything simpler nor more complicated...
Do you remove the asymmetry before testing?
chrisw
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
> mclane wrote:
>
> > Sorry but I am not capable of reading your mixed statements, quotes
> > and all this stuff.
> >
> > >HOW TO TEST FOR ASYMMETRY
> > >"Set up a position, and then let the program search for N plies, then
> >
> > >set
> > >it up again, play the move the program chose and let it search from
> > the
> > >other side for N-1 plies, and compare the evaluations..."
> >
> > Play a game of chess with program a and write down evaluations and
> > main-lines.
> > A has e.g. white.
> > After the game:
> > Setup the beginning-position but let black have the first move.
> > Let program a
> > now make the first black move. Try to replay the whole game with
> > inverse colores.
> >
> > If the scores or the main-lines are much different ....
> >
> > You can use games or the bt2630i for invers.
> >
>
> Chris's program is such an exception to the rule that this discussion
> is moot. Don't worry about it.
Good cop-out ! Must remember that style when arguing with my wife :)
> BTW your test for asymmetry doesn't accomplish a thing. If CSTAL's
> scores were different in your test, I would really start to worry for
> it.
Better start worrying, KK :)
> Your 2nd game is only a mirror image of the first. The first London
> tournament was played like that, but there is no difference in score
> evaluation from mirror images.
Sure ? :)
Chris Whittington
chrisw
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
> chrisw wrote:
>
> >
> >
> > Which is this group of asymmetric programs that are so "notoriously
> > difficult to change" ?
> >
> > And before you answer KK, let me give it you:
> >
> > large programs tend to be difficult to change.
> >
> > badly written programs tend to be difficult to change.
> >
> > kludged, mod-ed and hacked programs tend to be difficult to change.
> >
> > asymmetry has nothing to do with it.
> >
> > Chris Whittington
> >
> > >
> > > bruce
> > >
Oh God, another stream of KKKKKKKKKKKKKKK ......
>
> If you have an evaluation function that has x number of variables with
> parameters, and you then add a global adjustment to the score of that
> evaluation function,
so far so good,
but is this 'global adjustment' a function of the score or an absolute
value ? this has consequences for the rest of your 'arguments'
> what you have done is overidden the individual
> parameters
depending on the type of 'global adjustment' it would appear that 'I' have
have either scaled the score from the original function or added some
absolute value to it
> and substituted another function on top of the previous one.
substituted ? modified might be better. Hard to see how addition of
something substitutes it, but never mind.
> Now you are 2 levels removed from the variables themselves.
What the fuck does this mean ?
> So when you
> make a change to one of the parameters on either level, how can you
> isolate the rest of the parameters when the function on the other level
> is affecting their score?
My brain is in a spin with this one. How can you torture the english
language this way ?
> Or is your asymmetric code done differently
> than this?
Sorry, I'm completely lost now. And, er, and what asymmetric code is this ?
Chris Whittington
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
: Robert Hyatt wrote:
: >
: >
: > In Crafty, all scoring is initially processed in a fully symmetric
: > manner.
: > Then, at the end, Crafty says "Hey, I'm playing white, so my king
: > safety
: > is more important than my opponents". This is a simple multiply done
: > at
: > the end of things. Up to this point, everything is the same for both
: > sides... I've even played with a symmetric Crafty, and all that was
: > needed
: > was to remove the asymmetric multiplier. So symmetric or asymmetric
: > doesn't
: > necessarily make anything simpler nor more complicated...
: Do you remove the asymmetry before testing?
No. When I have crafty giving analysis on the servers (such as for the
various GM tournaments) I go in and hand-remove the assymetry multipliers
(there are only two at present, king safety and blocked pawns) so that as
it flips from one side to the other the scores don't oscillate and make
everyone ask questions about why.
I've not found any successful ways to make Crafty "attack like mad" just
yet. When I tune it more aggressively, it wins some interesting games, but
it also loses more games, particularly against computers. It plays better
being defensive. It will break the position open, given a chance, but it
doesn't "plan" on doing so for many moves in advance...
There have been "aggressive" crafty's... many remember the 9.x series, where
for a few versions it was going ballistic pushing pawns on the opponent's king
position. Worked for a while, but eventually, after the newness wore off,
it started failing miserably... and became part of "history"... :)
Komputer Korner
Robert Hyatt wrote:
>
>
> No. When I have crafty giving analysis on the servers (such as for
> the
> various GM tournaments) I go in and hand-remove the assymetry
> multipliers
> (there are only two at present, king safety and blocked pawns) so that
> as
> it flips from one side to the other the scores don't oscillate and
> make
> everyone ask questions about why.
>
> I've not found any successful ways to make Crafty "attack like mad"
> just
> yet. When I tune it more aggressively, it wins some interesting
> games, but
> it also loses more games, particularly against computers. It plays
> better
> being defensive. It will break the position open, given a chance, but
> it
> doesn't "plan" on doing so for many moves in advance...
>
If you don't remove the asymmetry before testing, how can you know how
much the tuning has improved the problems that asymmetry was designed to
alleviate. The asymmetry will always be a permanent crutch for the
evaluation function that isn't good enough in the first place to stop
the problems that the asymmetry was put in to fix. The long range
problem with the asymmetry is when do you know that you don't need it
any more, if you always have it in while testing? Don't forget that
asymmetry also causes problems. It isn't a free fix.
Bill Newton
In article <01bc9f21$62566ea0$c308...@cpsoft.demon.co.uk>,
chrisw <chr...@cpsoft.demon.co.uk> writes
>> Now you are 2 levels removed from the variables themselves.
>
>What the fuck does this mean ?
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Just a little reminder Christopher as you seem to have resorted
more and more of late to swearing in public, I guess in a vain attempt
to give your comments more impact.
It doesn't work.
Why?
Well, although you swear repeatedly you are not capable of swearing
with elegance. You try hard and swear often but sadly you will never
master the art of swearing with refinement, propriety or judgement.
So why not pack it up?
BTW I understand from earlier postings of yours that you allow your
children to read this newsgroup.
Surely you dont consider it appropriate for them to see their Father
using foul language that he wouldn't dream of using in his own home.
Well I assume you wouldn't use foul language in your own home.
Would you?
Additionally it may help you to ponder on the fact that 99% of
postings to this newsgroup actually manage to express a point of
view without swearing.
So why not yours?
Come on Christopher, have confidence in your non swearing
vocabulary! You can do it, you CAN do it if you really try.
Cant you?
Know what I mean Christopher?
--
Bill Newton
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
: Robert Hyatt wrote:
I turn it off on occasion, but taking just king safety, I pretty well
know when I can get rid of it, and that point is a *long* way off. When
Crafty can look at a position and decide whose king is most likely to
be successfully attacked, if both are somewhat opened up. That is not
something I have seen anyone do yet. Starting with Genius as a case in
point. I'm not sure what he does, but whatever it is, it is *not* good
for players that like to attack.
I don't consider this a total "crutch". I know players (Karpov comes to
mind) that would rather defend and let you wreck your position attacking,
and then roll you in the endgame. I'd suggest that his evaluation is quite
asymmetric in that regard... just as players who like to attack (Orlov on
ICC is a case in point here)...
Tom Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
: If you don't remove the asymmetry before testing, how can you know how
: much the tuning has improved the problems that asymmetry was designed to
: alleviate. The asymmetry will always be a permanent crutch for the
: evaluation function that isn't good enough in the first place to stop
: the problems that the asymmetry was put in to fix. The long range
: problem with the asymmetry is when do you know that you don't need it
: any more, if you always have it in while testing? Don't forget that
: asymmetry also causes problems. It isn't a free fix.
If I recall correctly, Korner, you have absolutely no idea what's wrong
with asymmetry. You're pretty sure it causes technical problems due to
a post Vincent (of all people) made months and months ago. A shiny new
nickel says you still don't have any idea what he was talking about. Let's
face it, Vincent is a bad source to be getting any sort of information
from. So anyway, you're now posting without having the foggiest idea of
what's going on and speculating about programming aspects that you have
no experience with. I think my point can be summed up by telling you to
shut up and leave the programming to the programmers.
Cheers,
Tom
mclane
"chrisw" <chr...@cpsoft.demon.co.uk> wrote:
>> It is the job of the opponent to destroy these dreams.
>I don't want my dreams destroyed :(
Than program a version that wins paris ! :-)
>Chris Whittington
mclane
"chrisw" <chr...@cpsoft.demon.co.uk> wrote:
>> Now you are 2 levels removed from the variables themselves.
>What the fuck does this mean ?
I don't understand it either.
I thought he speaks fluently english ?!
>> So when you
>> make a change to one of the parameters on either level, how can you
>> isolate the rest of the parameters when the function on the other level
>> is affecting their score?
>My brain is in a spin with this one. How can you torture the english
>language this way ?
Aha - not only me is recognizing this. Maybe KK is a german trying to
speak english in kanada ?
Or an italian in kanada ? Or an english man in NY ?
>> Or is your asymmetric code done differently
>> than this?
>Sorry, I'm completely lost now. And, er, and what asymmetric code is this ?
Which asymmetric code ??
>Chris Whittington
mclane
Komputer Korner <kor...@netcom.ca> wrote:
What has asymmetrie to do with different (x) numbers of parameters and
functions together in one evaluation function ??
I don't understand your problem KK.
mclane
"chrisw" <chr...@cpsoft.demon.co.uk> wrote:
Take Karpov - Topalov . If the program does not find Rxe6, take back
one move and see if it sees Rxe6 with this try. If so, you have
GENIUS.
Sometimes looking in a mirror helps !!
BTW: CSTal is in Alice wonderland. So anything happens...
mclane
Komputer Korner <kor...@netcom.ca> wrote:
KK- my main problem is, you are using a word: asymmetrie. And you use
it often as if it is a law of the bible or newtons law of gravitation.
My problem is: Which asymmetrie in chris' program do you refer ????
I don't see it ! Chris does not see it himself, although he has
programmed it!
Do you refer on different evaluations depending on the case if the ply
is computer or opponent depth ?
Do you refer on a kind of different search-pruning (e.g. more pruning
when the ply is computer-depth , less pruning or no pruning when the
ply is opponents depth ---- or vice versa) ?
Do you refer on a kind of preprocessor, deciding with root position
WHICH algorithms to use, and therefore bringing asymmetrie in the
program ??
I don't know HOW you use this word.
Please define your word for me, cause it could have different
meanings/usages.
> If you don't remove the asymmetry before testing, how can you know how
>much the tuning has improved the problems that asymmetry was designed to
>alleviate.
We do not remove any part when we test. We want to test the program as
it is. If we test it castrated, how do we know how it works without
castration ? We would need to test it twice.
No - we test it like it is. If there is an asymmetrie (I don't think
there is, depending on your definition) - we don't remove it.
>The asymmetry will always be a permanent crutch for the
>evaluation function that isn't good enough in the first place to stop
>the problems that the asymmetry was put in to fix.
Wrong.
>The long range
>problem with the asymmetry is when do you know that you don't need it
>any more, if you always have it in while testing? Don't forget that
>asymmetry also causes problems. It isn't a free fix.
I don't know why you believe asymmetrie is a problem. Also I still
don't know what you understand under it.
mclane
Bill Newton <Bi...@notwen.demon.co.uk> wrote:
>In article <01bc9f21$62566ea0$c308...@cpsoft.demon.co.uk>,
>chrisw <chr...@cpsoft.demon.co.uk> writes
>>> Now you are 2 levels removed from the variables themselves.
>>
>>What the fuck does this mean ?
>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Hello Bill, I will try to answer you, because Chris is in holiday...
>Just a little reminder Christopher as you seem to have resorted
His name is Chris. I don't know why you call him christopher. I
thought the name is something somebody choses. So I think you should
call somebody by the name he wishes, not by the name other people wish
him to have. But maybe you have different ideas about what is
accurate.
>BTW I understand from earlier postings of yours that you allow your
>children to read this newsgroup.
>Surely you dont consider it appropriate for them to see their Father
>using foul language that he wouldn't dream of using in his own home.
First:
Chris, as all we straight people, uses the same language in his home,
in his newsgroup, when he is somewhere else in the world.
WE DON'T CHANGE OUR APPEARANCE because we are 1:1 persons.
Maybe you find this impolite, and not civilized. But we have same
negative words for your values. We call your values two-faced, hiding
yourself behind polite words, puritanical ideas ...
>Well I assume you wouldn't use foul language in your own home.
You are wrong.
>Would you?
Yes, he would ! And I would do the same. As we all do. The only person
using a language like nuns is you. And even nuns fuck from time to
time.
>Additionally it may help you to ponder on the fact that 99% of
>postings to this newsgroup actually manage to express a point of
>view without swearing.
And ? When posting without swearing is the right way to civilize
people, why has united states the highest cirminal rates and the most
people in prison of all states in the world. I mean, a polite and not
swearing language should lead to peaceful and cultural citicens, or ?
>So why not yours?
>Come on Christopher, have confidence in your non swearing
>vocabulary! You can do it, you CAN do it if you really try.
Why should he ? Language is just a vehicle to express yourself. Not an
exercise to be licensed to exist. A vehicle ! Nothing more.
Nobody gets criminal because he uses a criminal language. Nobody
develops into a criminal because he says shit/fuck/...
Maybe you mix up reason and result.
Do you really believe somebody having a cruel impolite language will
be a cruel impolite person ??
Ha - I have seen many many criminals using a 100% political correct
language... and I know that criminals have not been criminals because
they were thought a certain language, but because they lived in a
social-environment of poorness, discrimination and without any hope.
If you see criminals in a shadow of a city of scyscrapers, you have to
see the big buidings of the higher society taking the poor people away
the sun. The criminals live in the slums. And the slums are in the
shadow of the big scyscrapers of rich people.
Language is not the reason. It does not cause anything.
It is a result.
>Cant you?
>Know what I mean Christopher?
There is no christopher who could answer you. If you would not call ME
by MY name, I would not answer YOU.
If you want water and ask for fire, you don't get the right thing.
Maybe you try it with CHRIS instead of christopher?
>--
>Bill Newton
Tord Kallqvist Romstad
Komputer Korner <kor...@netcom.ca> writes:
> HOW TO TEST FOR ASYMMETRY
> "Set up a position, and then let the program search for N plies, then
> set
> it up again, play the move the program chose and let it search from the
> other side for N-1 plies, and compare the evaluations..."
Tord's list of reasons why KK's asymmetry test doesn't work, version 1
----------------------------------------------------------------------
1. If the program does any preprocessing, it's evaluation tables will
probably be different during the second search, because the root position
has changed.
2. Various search parameters could change depending on the search depth.
For instance, the null-move reduction factor could be increased at higher
search depth, an idea proposed in this newsgroup recently. The way search
extensions are used might also depend on the search depth.
3. A related problem: In a selective program, an n-ply search might consist
of 4 plies of full-width search followed by n-4 plies of selective search.
KK's test would obviously fail for such a program.
4. The hash table is likely to mess up the results for many programs.
The test would probably work for brute force programs which do not use
hash tables, preprocessing or hash tables.
Tord
Bill Newton
In article <5s2maj$9um$1...@steve.prima.ruhr.de>, mclane
<mcl...@prima.ruhr.de> writes
>Hello Bill, I will try to answer you, because Chris is in holiday...
>
>>Just a little reminder Christopher as you seem to have resorted
>
>His name is Chris. I don't know why you call him christopher.
Hiya Macca, I called him Christopher because 'Chris' is usually the
shortened version of 'Christopher' in England. No discourtesy
intended.
Incidentally 'my' name is shortened to Bill from William, so if you wish
to address me as William................feel free, tis but a trivial
matter.
>>Well I assume you wouldn't use foul language in your own home.
>
>You are wrong.
I picture you sitting round the dinner table with close friends and
rosy cheeked children. "Pass the effin salt" you say. " Get it your effin
self " remarks one of the rosy cheeks. " you cheeky little ***tard" is
your considered response......... Yep, a cosy night at home with Macca
and Co.
>The only person
>using a language like nuns is you. And even nuns fuck from time to
>time.
You know Macca, there's a time and a place for everything, even
swearing.
But as is becoming usual you misconstrue my posting. Believe me, I'm
no Angel and after a workout in the gym the air in the mens locker
room is often quite blue..... and in the Bar on a Saturday night.......!!!
But you surely understand that that is a world away from using such
language in posting to a newsgroup.
And to use it in your own home....... to Children..........Hmmmnnn. Sad.
Do you understand the word 'Decorum'? If you dont I would suggest
that you look it up and thoroughly digest its meaning. Because you
appear to lack it somewhat.
>If you want water and ask for fire, you don't get the right thing.
>Maybe you try it with CHRIS instead of christopher?
This is a riddle! Right?
Look forward to your considered comments.
Regards.
--
Bill Newton
mclane
Bill Newton <Bi...@notwen.demon.co.uk> wrote:
>Hiya Macca, I called him Christopher because 'Chris' is usually the
>shortened version of 'Christopher' in England. No discourtesy
>intended.
As I told you before, a name is not a name of somebody because society
WANTS to give a name. Any individual has the right to get his own name
spoken out the way he wishes.
If you have nothing against people calling you William, thats your
problem. But I think Chris wants to be named CHRIS.
Names are not for the others, they are for the persons who get this
name.
>Incidentally 'my' name is shortened to Bill from William, so if you wish
>to address me as William................feel free, tis but a trivial
>matter.
>>>Well I assume you wouldn't use foul language in your own home.
>>
>>You are wrong.
>I picture you sitting round the dinner table with close friends and
>rosy cheeked children. "Pass the effin salt" you say. " Get it your effin
>self " remarks one of the rosy cheeks. " you cheeky little ***tard" is
>your considered response......... Yep, a cosy night at home with Macca
>and Co.
What is Macca and who is Macca ? Which language is Macca ?
I don't understand this word. If it should be german, you write it
wrong.
>
>>The only person
>>using a language like nuns is you. And even nuns fuck from time to
>>time.
>You know Macca, there's a time and a place for everything, even
>swearing.
Yes.
>But as is becoming usual you misconstrue my posting. Believe me, I'm
>no Angel and after a workout in the gym the air in the mens locker
>room is often quite blue..... and in the Bar on a Saturday night.......!!!
Thats right.
>But you surely understand that that is a world away from using such
>language in posting to a newsgroup.
Aha. Why ?
>And to use it in your own home....... to Children..........Hmmmnnn. Sad.
Why ? I prefer using dirty language and not raping/misusing my
children instead of beeing a prof. and using good language and
raping/misusing my children...
>Do you understand the word 'Decorum'? If you dont I would suggest
>that you look it up and thoroughly digest its meaning. Because you
>appear to lack it somewhat.
Decorum is when I lend you 1 DM and you give it back to me.
Decorum is when Jaap van den Herik says: Come with me, or I 'll kick
you in the face.
>>If you want water and ask for fire, you don't get the right thing.
>>Maybe you try it with CHRIS instead of christopher?
>This is a riddle! Right?
No - it is a trivial use of language. The first lesson:
If you are thursty, ask for water. But don't call for fire.
WATER means WATER, not fire. If you call water fire, you will get
wrong thing.
Bill Newton
In article <5s5ibp$pka$2...@steve.prima.ruhr.de>, mclane
<mcl...@prima.ruhr.de> writes
>As I told you before, a name is not a name of somebody because
>society
>WANTS to give a name. Any individual has the right to get his own
>name
>spoken out the way he wishes.
>If you have nothing against people calling you William, thats your
>problem. But I think Chris wants to be named CHRIS.
>Names are not for the others, they are for the persons who get
>this
>name.
>>I picture you sitting round the dinner table with close friends and
>>rosy cheeked children. "Pass the effin salt" you say. " Get it your
>>effin
>>self " remarks one of the rosy cheeks. " you cheeky little ***tard"
>>is
>>your considered response......... Yep, a cosy night at home with
>>Macca
>>and Co.
>
>What is Macca and who is Macca ? Which language is Macca ?
>I don't understand this word. If it should be german, you write it
>wrong.
I've explained to you before that 'Macca' is another way of referring
to 'anybody' whose surname begins with Mc, like in mclane. In other
words 'Macca' is 'mclane' and 'mclane' is 'you' Gottit yet!?
>>And to use it in your own home....... to Children..........Hmmmnnn.
>>Sad.
>
>Why ? I prefer using dirty language and not raping/misusing my
>children instead of beeing a prof. and using good language and
>raping/misusing my children...
What peculiar principles you have. Does it not occur to you
that many children are happily raised, without being abused, and
without being subjected to obscene language in their own home?
Your own peculiar logic suggests that 'you' prefer to use 'dirty'
language as a preventative measure enabling you to avoid abusing
your own children.
If what you write is true, you may well have a serious medical
problem, in which case I would urge you to seek Professional
guidance.
>>Do you understand the word 'Decorum'? If you dont I would
>>suggest
>>that you look it up and thoroughly digest its meaning. Because you
>>appear to lack it somewhat.
>
>Decorum is when I lend you 1 DM and you give it back to me.
>Decorum is when Jaap van den Herik says: Come with me, or I 'll
>kick
>you in the face.
Come on Macca, admit it.; you dont understand it do you? Back to
your lexicon my friend and have another go:)
>>>If you want water and ask for fire, you don't get the right thing.
>>>Maybe you try it with CHRIS instead of christopher?
>
>>This is a riddle! Right?
>
>No - it is a trivial use of language. The first lesson:
>If you are thursty, ask for water. But don't call for fire.
>WATER means WATER, not fire. If you call water fire, you will get
>wrong thing.
This is another riddle! Right?
Regards.
--
Bill Newton
mclane
Bill Newton <Bi...@notwen.demon.co.uk> wrote:
>>You are right Rolf, it is not important. Only a name... :-)
>LOL! You really think I'm Rolf?
No - but I will call you Rolf in all future posts because it fits
better to my pattern-system. Rolf is Rolf is Rolf.
>Tell you something funny. It wasn't too long ago that your
>countryman Rolf wrote to this newsgroup asserting that 'I' was in
>fact Chris Whittington!!
>Strange folk you Germans.
Yes - thats true. We are strange. Democracy is something we have never
learned. It has developed into a moarchical system here, where the old
power sources (IG-Farben, Dresdner Bank, Deutsche Bank, Henkel, BASF,
the big energy-companies) have not been destroyed. They took a new
name, or still use the name of the jews they killed, and used their
money, and have still the power about anything. Nazi's have not
changed. The judgement of Nuernberg was not consequent, and Kohl shut
down germany into a country of 6-7.000.000 unemployees. 10 years ago
we had 2-3.000.000. But he started his game 1982, so he was very good
in misuing the country. But germans vote for him, cause he looks like
a strong leader. Germans still like strong leaders.
>>shall be the reason for beeing criminal is somehow weird.
>That would indeed be a very weird conclusion!
So why do you insist using polite and 2 faced language ?
>A joke!!!??? Come on..........Germans don't joke......it's not in their
>culture!
Right too. We can only laugh when we are drunken with german beer. And
when we are drunken we sing german folk-songs.
Horrible !
>>I don't have children!
>Then why did you create the impression that you do have children by
>recently posting the following:
It was something to be said in general.
>Believe me, 'Rolf' is wrong is wrong is wrong! LOL!
Yes - I can subscribe to this statement too.
You see - more and more we find a language to communicate...
>Tell you what.......check it out with Chris. He'll spot the difference in
>writing styles without trying. Now Rolf's English is good, but as a
>born and bred Brit I do have the edge on him, honest! Ask Chris or
>any Brit, they'll put you right.
Right. Here you call him Chris. Good done Bill.
>But you have put a smile on my face :)
Andrew Tridgell
I use quite a few asymmetric factors in KnightCap, in both the search
and evaluation. The reason isn't to steer the position into things
that the computer handles better, its just to play better chess
and to reduce the node count.
In particular some asymmetries I use are:
Search asymmetry: use search extensions only when KnightCap is in
trouble not the opponent. Thus I don't extend on check
or singularities if it is the opponent who is in check.
Similarly I only extend on pawn pushes to the 6th and 7th
ranks if it is the opponens pawn that is being pushed.
The reason for this search asymmetry is that it reduces
the node count a lot (due to half the extensions) and
it actually plays better moves. The idea of search extensions
(at least in my opinion) is to stop the computer from
playing blunders. Using search extenions to try to find
deep mates and tactics is just a waste of time against
a good opponent, as nearly all the time you don't find
anything. Sure, it can make you solve more of those
tactical test suites, but I'd much rather be able
to refute tactics by the opponent than spot the occasional
brilliancy.
Eval asymmetry: I make uncertain factors asymmetric. The more
uncertain a factor is the more asymmetric is should be. The
classic example is king safety. Having king safety symmetric
means that you are saying that "if I have rotten king safety
but so does the opponent then we're even". Thats extremely
dangerous, and leads to losses against opponents that can
better estimate king safety (ie. most humans).
Similarly for things like trapped pieces. You want to avoid
trapping one of your own pieces, but trapping an opponents
piece isn't worth nearly as much, because otherwise you'll
give up material to trap it, and then may find the trap
doesn't hold. Thats an easy way to lose to opponents that
evaluate trapped pieces more accurately than you.
The whole issue of asymmetry can be summed up as "if you
really want to win then unclear positions are bad".
Cheers, Andrew
Bill Newton
In article <5srumk$7...@drn.zippo.com>, Graham Douglass <?@?.?>
writes
>I sincerely hope to avoid getting dragged into an argument about this, so I
>may
>not reply to any replies, but I think the underlying reason why Chris uses
>"strong" language is that he is a control freak, but he possibly doesn't feel
>he's fully in control of events at the moment.
A concise and credible observation that makes your point of view
crystal clear. All in respectable language too! Admirable stuff!
Regards.
--
Bill Newton