The end of computer chess progress?
Ed Schroder
>Search (simple material chess evaluation over a tree) will lead to
emergent
>properties. Some chess knowledge will emerge quickly (like don't leave
your
>king and queen in a knight fork position). Some will emerge slowly
(maybe
>like your queen development). And some effectively never. Where
quickly,
>slowly and never refer to the size of the search space.
[ snip ]
>The fact that this process of fire control doesn't work is masked by
the
>fact that programs keep getting faster. More generalised encapsulations
of
>higher-level knowledge are needed - that's where I spend (some of) my
time.
>
>Chris Whittington
Chris,
What do you think about the statement, "Chess in 98% tactics", this
limited to computer-computer games area, let me explain a bit...
I remember the (old) Hitech experiment of Berliner. He took the Hitech
program removed important chess knowledge but left the very basic
chess knowledge and called this version Hitech_Low (HL) and his
original version HH.
Then matches were played with HL and HH at several ply-depths.
It showed up that HL (8-ply) lost from HH (8-ply) but that HL (9-ply)
vs HH (8-ply) was about equal and HH (8-ply) was crushed when
it played HL (10-ply).
The conclusion more or less was: (extra) chess knowledge is just
worth one ply.
Especially the last years I have seen the following happen:
. make the program as fast as possible (NPS)
. this in combination with a low branch factor (null-move)
. add only the very basic chess knowledge (piece-square tables etc)
. make the program aggressive to ensure tactic complications in games.
As a result (also because of today's fast Pc's) you get your program
on incredible deep ply-depths. 13 and 14 plies in the middle game are
no exceptions anymore.
What if a program with a lot of chess knowledge (having a low
NPS because of that) is faced with such a "fast searcher" AND the
HH_HL principle is true. It will lose no? The "fast searcher" naturally
comes 3-4 plies deeper because of a smart search and the HH_HL
principle ensures victory after victory against a low NPS program.
True sofar?
What do you think of the following statement of mine:
In comp-comp a smart search is worth 200-300 ELO points.
IMO many programmers have done the following the last years:
- remove (!!) chess knowledge from the program to get a high NPS;
- improved selected search (smart pruning) to get a low branch factor;
- make the program more aggressive to ensure tactics in comp-comp games;
As a result the programs do a lot better now in the comp-comp area.
Is this the end of adding new chess knowledge to our programs?
It bothers me. To compete in the comp-comp area you have to
remove chess knowledge???
The bad news is, it apparently seems to work!
This subject has been my attention since a couple of months now.
It's not that simple as you have for instance Hiarcs and Chessmaster
which are "slow searchers" (a very low NPS) and they both are doing
fine in comp-comp.
However if you take a closer look both programs seems to be
specialized in tactics unless the fact they have a low NPS.
The last months I have worked on Rebel adding tactics, more pruning,
a lower branch factor. As a result it performs better in comp-comp but
I definitely see a drop in the positional play. The quality of the
returned
moves is simply less, still it is a better comp-comp fighter now.
What to do?
Every time a new program is released the program is judged in
magazines (or internet) because it is compared with other chess
programs and the score it gained against them.
To compete in this area you have to make your program a weaker
positional player????
And the fact that important human-comp tournaments (AEGON,
Harvard cup) are gone makes comp-comp even more important
or even worse you can ask yourself the question: what else is left?
Thinking all of this over the last months I came more or less to the
conclusion that we have entered a new area, that adding new extra
chess knowledge works out counter productive and that progress
in computer chess will stop or already has been stopped as I can
not believe that removing chess knowledge is progress.
The fact that Program_X scores better against Program_Y doesn't
(necessarily) mean that X is better than Y. X just scores better
against Y and that's all. Maybe Y scores a lot better against humans
than X and Aegon certainly has proven that.
So where do we stand now in computer chess?
And more important, what is the way to go?
Ed Schroder
ShaktiFire
the way to go is to stop worrying
about better ratings, and instead
include some coaching/learning
features. Improving the engine,
when the computer beats 99% of
the players anyhow, is not necessary.
Actually I enjoy a program that plays
a good move, a good positional game,
a solid postion better than a program
that can outcalculate a player in wild
tactical postions. I guess I would argue,
from the consumer viewpoint, that it is
better to have a program thats plays
"beautiful" chess, then one that can
get high computer-computer ratings.
Somehow, I think those computer-computer
ratings are a bunch of nonsense. I think the
opening book ccoks and pure nps tactical
schemes to simply get higher computer-computer ratings detract from
"quality" play as judged by human standards.
In some sense, ratings are a game you
programmers and computer chess hobbyists
enjoy. But in many ways these SSDF ratings
are a sham. You programmers should know
that most of all.
en...@intercom.es
This applies to Chessmaster, but not to Hiarcs 7, and it is an important
exception to the general rule you mention. Hiarcs is by far the slowest NPS
and does great in comp-comp games at any speed, including blitz. I tried H7
lowering its selectivity from 5 (default) to 2, making it less tactical and
more knowledge based, and it still does great. Is this enough to dismantle
your theory?
It is true that programs that do best in comp-comp games are very aggressive,
to the point that their performance in my tournament was directly
proportional to aggressiveness. But it is wrong to say that this is important
“to ensure tactics in comp-comp games”. If this were the case, Nimzo 99, as
much of a fast finder as anything, would take advantage from the
aggressiveness of the opponents, and it doesn’t because it is rather passive
and the tactical positions are built against it. The best programs take
advantage of their aggressiveness by building up positions that favor them,
very much like human grandmasters do. There is no such a thing as a passive
GM. They may be positional, quiet, a la Karpov, but never passive, and some
programs are and pay for it. Not knowing enough to play quiet and long range
plans, the best programs play aggressive and create positions that give them
more chances than to the opponent. You can consider this characteristic as
pre-tactical, as an “intelligent” way to make them play, applied today by
some slow and fast searchers. So you could say that being aggressive is
important “to ensure tactics in comp-comp games after creating a favorable
position”, which is a manifestation of knowledge.
I read statements made by Donninger and Théron claiming that they wanted to
get rid of “unnecessary” knowledge, and even not being a programmer it makes
an awful lot of sense to me. It is clear that knowledge has a different role
now at 400K N/S than 16 years ago when Mephisto III ran at 2 nodes per second
and the poor thing had to compensate for it. Maybe we are not talking about
further development of knowledge, but about a redefinition of what kind of
knowledge matters more than the search being slowed down by it. Alright, I
already said I am no programmer, but this is how things look like to me from
the "“outside"”
As for the drop of quality in positional play, it is not obvious to me
either. Aside from the already mentioned Hiarcs 7, Junior 5 and Tiger play a
quality game that I never saw before in fast searchers, and very rare in slow
searchers even today.
By the way, the lowering of the branching factor doesn’t necessarily apply
either as a common trend in new engines. For instance, Junior 5, in my opinion
one of the two very best engines around, has the lowest of all together with
Genius, even lower than Hiarcs and much lower than Rebel.
Enrique
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Chris Whittington
Ed Schroder wrote in message <36E128CC...@rebel.nl>...
Er ...... this is *the* Ed Schroder talking to me ?
Yup, checked the header file. It is he.
Why this ? From the one who played such a large part in the unwarranted
censorship of my goodself for all that time ?
Anyway, polite as I am, I shall endeavour to reply to your post ....
>>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>
>>Search (simple material chess evaluation over a tree) will lead to
>emergent
>>properties. Some chess knowledge will emerge quickly (like don't leave
>your
>>king and queen in a knight fork position). Some will emerge slowly
>(maybe
>>like your queen development). And some effectively never. Where
>quickly,
>>slowly and never refer to the size of the search space.
>
>[ snip ]
>
>>The fact that this process of fire control doesn't work is masked by
>the
>>fact that programs keep getting faster. More generalised encapsulations
>of
>>higher-level knowledge are needed - that's where I spend (some of) my
>time.
>>
>>Chris Whittington
>
>Chris,
>
>What do you think about the statement, "Chess in 98% tactics", this
>limited to computer-computer games area, let me explain a bit...
>
>I remember the (old) Hitech experiment of Berliner. He took the Hitech
>program removed important chess knowledge but left the very basic
>chess knowledge and called this version Hitech_Low (HL) and his
>original version HH.
>
>Then matches were played with HL and HH at several ply-depths.
>
>It showed up that HL (8-ply) lost from HH (8-ply) but that HL (9-ply)
>vs HH (8-ply) was about equal and HH (8-ply) was crushed when
>it played HL (10-ply).
>
>The conclusion more or less was: (extra) chess knowledge is just
>worth one ply.
Yup, yup, yup. Actually the conclusion is that *Berliner's* encoded
knowledge (HH minus HL) was worth 1 ply. I think he wanted to work up the
argument that Botwinnik's knowledge was therefore also only worth one ply.
Note how nobody ever questioned the knowledge. it came from the great man,
so it must have been *the_knowledge*, innit ?
Anyway, such is the power of the academic research nonsense that it goes
down in history for bean-counter justification regurgification whenever
necessary.
Both you and I know that (value of knowledge > 1 ply). Depends on the
knowledge and the position.
>
>Especially the last years I have seen the following happen:
>. make the program as fast as possible (NPS)
>. this in combination with a low branch factor (null-move)
>. add only the very basic chess knowledge (piece-square tables etc)
>. make the program aggressive to ensure tactic complications in games.
>
Yup. yup, yup.
Boring isn't it ?
And it's so easy. Which is why there are so many programs. I work at the
moment with an assistant programmer. He can play chess, which helps. In the
old days, I used to think chess programming was a huge complex process, but
that was because I used to do everything, user interface, QA, bug testing
blah-blah-blah. Now, with second programmer, I zap out the code, and he (and
me too if necessary) clean it up and make sure it works. This way, I keep
the overview, and avoid getting bogged down in the detail.
It took us two months to write the nitty-grittys of a bean-counter. That was
the start point of the new project, then to go onto higher things. Based on
better than bitmaps. Similar idea, but better.
Then, to go bean-counter, just a bit of eval, and the usual tricks. Bob is
your uncle.
The bad news is that any drongo can do this. Worse, any drongo does do this.
Once I used to think chess programmers had something special. Therefore
their minds were high quality. Unfortunately, listening to them on the news
groups, with words, ideas, stuff that provokes thought ... there is none.
Chess programmers are dorks. Simple but true. If they had brains, they'ld
show it with words, understanding and ideas. Only they don't. Look to the
top and what do you see ? An idiot.
So what we got ? A whole load of drongos who easily as hell can put
bean-counters together, and these bean-counters can play. Play shit style
but play.
>
>As a result (also because of today's fast Pc's) you get your program
>on incredible deep ply-depths. 13 and 14 plies in the middle game are
>no exceptions anymore.
Right. My first bean-counter (I just described it), I had it playing 5 ply,
full width, piece square tables only, no pruning. Came back with move
immediate. Just to see if it worked. All connetced to the CSTal WIn 2
interface. The bloody thing smashed me.
CSTal2 doesn't smash me, it plays positional, I can understand the game,
there are themes, it's consistent. Probably Rebel is like this for you too.
If I lose, its because I'm ground down, I can see why and how. But this
5-ply thing, I had no respect for, I just played my usual positional stuff,
not really thinking, and it random killed me. Not because it played for it,
not because of a theme, not for any understandable chess reason. Just a
random minefield. Oh, whoops, I trod on another mine.
This is not chess, it's just stupid pointless nonsense. Might as well toss a
coin or play Russian roulette.
This is how it is a 5 ply, with me not bothering to think. It's the same
random minefield at 15 ply. Only the fools don't realise. They don't realise
because they can't play chess. Can't play chess, buy a bean-counter -
simple. And they do.
>
>What if a program with a lot of chess knowledge (having a low
>NPS because of that) is faced with such a "fast searcher" AND the
>HH_HL principle is true. It will lose no? The "fast searcher" naturally
>comes 3-4 plies deeper because of a smart search and the HH_HL
>principle ensures victory after victory against a low NPS program.
>
>True sofar?
No.
What is true is, what is the point of trying to make knowledge program that
can play with style, themes, and chessisms, when the masses don't have the
brainpower to understand it, and hordes of dork programmers can knock up
competing bean-counters in a few months ?
>
>What do you think of the following statement of mine:
>
>In comp-comp a smart search is worth 200-300 ELO points.
Right. In the land of the blind, having a longer feely stick works wonders.
>
>IMO many programmers have done the following the last years:
>- remove (!!) chess knowledge from the program to get a high NPS;
Right. Bastards.
>- improved selected search (smart pruning) to get a low branch factor;
>- make the program more aggressive to ensure tactics in comp-comp games;
>
>As a result the programs do a lot better now in the comp-comp area.
Yup. Boring isn't it ?
>
>Is this the end of adding new chess knowledge to our programs?
Nah. You just got to get the dynamic knowledge encoded. How do you think
CSTal copes at 4 knps against 300knps ? Or how about CSTal at 4knps to get
the good positions, then switch to bean-counter CSTal at 300 knps to finish
them off ?
It gets a lot of win positions that it then can't convert right now.
The dorks are frightened, believe me.
>
>It bothers me. To compete in the comp-comp area you have to
>remove chess knowledge???
>
>The bad news is, it apparently seems to work!
Sure. If nobody can see anything, then may as well cut away every redundant
thing, and rely on the feely stick.
Only needs one man with an eye and they are all kaput.
>
>This subject has been my attention since a couple of months now.
>It's not that simple as you have for instance Hiarcs and Chessmaster
>which are "slow searchers" (a very low NPS) and they both are doing
>fine in comp-comp.
>
>However if you take a closer look both programs seems to be
>specialized in tactics unless the fact they have a low NPS.
>
>The last months I have worked on Rebel adding tactics, more pruning,
>a lower branch factor. As a result it performs better in comp-comp but
>I definitely see a drop in the positional play. The quality of the
>returned
>moves is simply less, still it is a better comp-comp fighter now.
>
>What to do?
Dunno ? Do you care that much ?
I know I can make something that will wipe them out. Really now a question
of whether I can be bothered to do it.
>
>Every time a new program is released the program is judged in
>magazines (or internet) because it is compared with other chess
>programs and the score it gained against them.
Stupid. Stupid. Too many stupids. And they have a tendency to lie as well.
>
>To compete in this area you have to make your program a weaker
>positional player????
>
>And the fact that important human-comp tournaments (AEGON,
>Harvard cup) are gone makes comp-comp even more important
>or even worse you can ask yourself the question: what else is left?
Shall we jump ?
.
.
.
.
.
..
Splat :)
>
>Thinking all of this over the last months I came more or less to the
>conclusion that we have entered a new area, that adding new extra
>chess knowledge works out counter productive and that progress
>in computer chess will stop or already has been stopped as I can
>not believe that removing chess knowledge is progress.
Its that bloody Hyat, isn't it ? With his one paradigm, and his shouting the
one and only way, and his driving everybody else off the ng by answering
everything, Crafty this, Crafty that, Crafty the bloody other.
So now he's convinced you THERE IS ONLY ONE WAY ?
Czub - Wo bist du ?
>
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
>
>So where do we stand now in computer chess?
Well, I try and diversify away from from it. Too many stupid people. Maybe I
should defeat them all and then swan off ?
>
>And more important, what is the way to go?
Upwards, Ed.
And away from Hyatt. And the Great Satan.
Chris Whittington
>
>Ed Schroder
>
>
Ernst A. Heinz
> [...]
>
> Actually the conclusion is that *Berliner's* encoded
> knowledge (HH minus HL) was worth 1 ply. I think he wanted to work up the
> argument that Botwinnik's knowledge was therefore also only worth one ply.
>
> [...]
Actually, you cannot conclude anything from Berliner et al.'s
publication with *good confidence* because the presented data is
not statistically significant. Most of their matches (with just
16 games per match) feature such high margins of statistical error
that they do not even allow to discriminate between stronger and
weaker opponents with 90% confidence.
=Ernst=
Robert Hyatt
:>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
: [ snip ]
: Chris,
: Especially the last years I have seen the following happen:
: . make the program as fast as possible (NPS)
: . this in combination with a low branch factor (null-move)
: . add only the very basic chess knowledge (piece-square tables etc)
: . make the program aggressive to ensure tactic complications in games.
: As a result (also because of today's fast Pc's) you get your program
: on incredible deep ply-depths. 13 and 14 plies in the middle game are
: no exceptions anymore.
: What if a program with a lot of chess knowledge (having a low
: NPS because of that) is faced with such a "fast searcher" AND the
: HH_HL principle is true. It will lose no? The "fast searcher" naturally
: comes 3-4 plies deeper because of a smart search and the HH_HL
: principle ensures victory after victory against a low NPS program.
: True sofar?
: What do you think of the following statement of mine:
: In comp-comp a smart search is worth 200-300 ELO points.
: IMO many programmers have done the following the last years:
: - remove (!!) chess knowledge from the program to get a high NPS;
: - improved selected search (smart pruning) to get a low branch factor;
: - make the program more aggressive to ensure tactics in comp-comp games;
: As a result the programs do a lot better now in the comp-comp area.
: Is this the end of adding new chess knowledge to our programs?
: It bothers me. To compete in the comp-comp area you have to
: remove chess knowledge???
: The bad news is, it apparently seems to work!
: This subject has been my attention since a couple of months now.
: It's not that simple as you have for instance Hiarcs and Chessmaster
: which are "slow searchers" (a very low NPS) and they both are doing
: fine in comp-comp.
: However if you take a closer look both programs seems to be
: specialized in tactics unless the fact they have a low NPS.
: The last months I have worked on Rebel adding tactics, more pruning,
: a lower branch factor. As a result it performs better in comp-comp but
: I definitely see a drop in the positional play. The quality of the
: returned
: moves is simply less, still it is a better comp-comp fighter now.
: What to do?
: Every time a new program is released the program is judged in
: magazines (or internet) because it is compared with other chess
: programs and the score it gained against them.
: To compete in this area you have to make your program a weaker
: positional player????
: And the fact that important human-comp tournaments (AEGON,
: Harvard cup) are gone makes comp-comp even more important
: or even worse you can ask yourself the question: what else is left?
: Thinking all of this over the last months I came more or less to the
: conclusion that we have entered a new area, that adding new extra
: chess knowledge works out counter productive and that progress
: in computer chess will stop or already has been stopped as I can
: not believe that removing chess knowledge is progress.
: The fact that Program_X scores better against Program_Y doesn't
: (necessarily) mean that X is better than Y. X just scores better
: against Y and that's all. Maybe Y scores a lot better against humans
: than X and Aegon certainly has proven that.
: So where do we stand now in computer chess?
: And more important, what is the way to go?
: Ed Schroder
There is an alternative. Do as I try to do... Ignore computer vs
computer games unless you see something you are consistently doing wrong
in those games. I'm not hung up on trying to beat every program by
being as fast as possible. I have gotten _slower_ the past 3 years, not
faster, and I think that direction is perfectly ok. I've said before
that _far_ too much attention and emphasis has been placed on SSDF
results, which has led to today's situation. I think it much more interesting
to continue what I've been doing for along time: trying to find ways to make
these 'anti-computer' strategies backfire. Because once you learn how to
'break anti-computer' you also learn how to _play_ it. And there would be
nothing to prevent you from using that against other computers, of course. :)
The hardest problem I am looking at is the book. It is hard to realize that
when you play move 2 and 3 in some opening, that you are committing to either
locking the pawns up 10 moves further along, or else you are going to really
have to compromise your position to avoid that. And I don't see anyone able
to handle this yet, other than by hand-tuning. I have 'partially' solved the
inability to understand locked up pawns... but only partially, and I still
have the problem of committing to a locked pawn structure without knowing it,
because of the book line chosen.
BTW, note that "LowTech" in Berliner's paper was not _stupid_. It still
did a _full_ pawn structure evaluation, but after that only had a weak
centralize-the-pieces eval. However, it _might_ be that the pawn structure
analysis was the most important thing he did. A poorly placed knight can
be moved, while a weak pawn remains weak forever.
His results might be a little harder to quantify when you factor that in.
--
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
Rjpawlak
>the way to go is to stop worrying
>about better ratings, and instead
>include some coaching/learning
>features. Improving the engine,
>when the computer beats 99% of
>the players anyhow, is not necessary.
>
Very true. However, in the absence of other figures of merit for judging
whether a program is good or not, people tend to become a little obsessed with
rating numbers.
>In some sense, ratings are a game you
>programmers and computer chess hobbyists
>enjoy. But in many ways these SSDF ratings
>are a sham. You programmers should know
>that most of all.
I think that this 'game' is played because the market drives it. I see many
posts asking "Which program is the Strongest?" In many cases that is the wrong
questions, and the person asking it never realizes that fact.
What I would suggest is an alternative set of benchmarks. Compile a collection
of games played by GMs with a solid playing style (that contain no blunders by
the player in question). Then, use the analysis modes of each program to
determine a percentage agreement (at some time control). That would give people
another set of metrics to use when comparing programs.
Right now, there is only one 'number'/metric that people can point to when
evaluating these programs. They need another measure that is easily
quantifiable and understandable.
Clearly a becnhmark like this would also be used by the comp cheats, in an
effort to avoid detection. I don't have a solution for this problem....
Bob P. (Alias MLK and RJP - newsfeed is down)
Chess Widower's Home Page
Chess Software Reviews
http://members.aol.com/rjpawlak/ch_widow.html
Rolf Tueschen
>>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>[ snip ]
>Chris,
>True sofar?
>What to do?
Excuse me that I dare to enter such a typical expert question. But I
liked your post for a very personal reason. I think that human v comp/
comp v human playing is much more important. And we should try to
reanimate such tournaments.
You wrote about a real paradoxon. There might be a prog that wins
against all other progs and still -- it's not necessarily the
best/strongest 'player'.
Ed, as I could see more than once, you are the guy to turn around the
wheel if you once was convicted that the direction was wrong.
I tell you what. Forget about all the tuning over-night with autoplayer.
Simply forget it. Have the guts to simply deny to do it. Let the rest do
it. And I for one promise you one thing. If you can show that your
approach leads you to better results in games vs, humans you will be the
new leader of the pack. The actual and past tradition of imbreeding in
comp v comp is nuts Ed, completely inhuman nonsense. And the clients
don't want the REBEL who beats them all (machine-wise) but they want a
hell of machine that is able to play human beings. Totally different
thing.
Methinks that even Bob there is more advanced than you commercial guys.
Ed, you are not forced to repeat the old tradition on and on. You even
don't need sort of contract with your competition. Just do it. And in
business talkie talk, _you_ are the new leader of a new trend. Period.
All this beyond all the bad past between us. Good to see you back here.
No joking or irony intended. I always said that you belonged into an
open forum. :)
>Ed Schroder
Ed Schroder
>>The last months I have worked on Rebel adding tactics, more pruning,
>>a lower branch factor. As a result it performs better in comp-comp but
>>I definitely see a drop in the positional play. The quality of the
>>returned moves is simply less, still it is a better comp-comp fighter now.
>>
>>What to do?
>
>Dunno ? Do you care that much ?
Yes I do.
I care because the playing strength of programs is judged by playing
against other programs also by running test-sets comparing solution
times of key moves.
So in order to get a positive review your program need to perform
well in comp-comp and in these test-sets.
As a result a programmer tends to add even more tactics. More of
the stuff it already excels in. Version 1.0 was a tactical monster of
2600, version 2.0 being 2800 now but in positional play version 1.0
and 2.0 are equal and maybe 2.0 is even weaker but since it scores
better in comp-comp and test-sets version 2.0 is called stronger.
Here is a snip from my home page from an interview I had with
Rochade Europa (march 1999). It gives a small example of this.
[ BEGIN ]
ES: I have mixed feelings about this. In my opinion human-comp
is much more important than comp-comp after all the program
is written for humans. But since the big human-comp events
have disappeared (AEGON, Harvard Cup) what else is left other
than to play comp-comp? So comp-comp becomes even more
important.
It's my opinion this indirectly (and unwanted) hurts the progress
in the quality of the chess engines as programmers are forced
to concentrate on good comp-comp results which I consider a
very bad development as being the (suggested) strongest
comp-comp program doesn't necessarily mean you are the
strongest against humans too (and AEGON has proved that)
and because programmers are more or less forced to focus
on good comp-comp results the quality of the returned moves
by chess programs may even lower.
One (small) example, Rebel since version 9 is able to find forced
checkmate combinations up to checkmates in 30 moves.
5n2/B3K3/2p2Np1/4k3/7P/3bN1P1/2Prn1P1/1q6 w - - bm e3g4;
If you import the following position into Rebel9 or 10 the program
will find the "mate in 30 moves" within one second. In principal this
is crazy, you don't need to search that deep against humans as
the computer by far is superior in these kind of positions. It would
be an improvement to limit these kind of deep searches to (say)
"mate in 12 moves" against humans with as result the program
will be faster (so better) in the overall performance against humans.
But the code will not be changed for reasons that other chess
programs do the very same (deep tactics) so you simply need
it if you compete in the comp-comp area.
[ END ]
Meaning to say, by taking out some tactics, limit extensions I can make
my program a better human-comp player. But you don't do that because
that would hurt in the comp-comp area.
This is just one (small) example how comp-comp works out counter
productive and stops real progress as I have called it in the header
and is my only reason why I have started this subject.
>I know I can make something that will wipe them out. Really now a question
>of whether I can be bothered to do it.
I think your idea to combine CST (4000) and CST (300,000) sounds
convincing enough to give it a try.
>>Every time a new program is released the program is judged in
>>magazines (or internet) because it is compared with other chess
>>programs and the score it gained against them.
>
>Stupid. Stupid. Too many stupids. And they have a tendency to lie as well.
>
>>
>>To compete in this area you have to make your program a weaker
>>positional player????
>>
>>And the fact that important human-comp tournaments (AEGON,
>>Harvard cup) are gone makes comp-comp even more important
>>or even worse you can ask yourself the question: what else is left?
>
>Shall we jump ?
>
>.
>.
>.
>.
>.
>..
>Splat :)
Mayday, mayday.... still wet :)
>>Thinking all of this over the last months I came more or less to the
>>conclusion that we have entered a new area, that adding new extra
>>chess knowledge works out counter productive and that progress
>>in computer chess will stop or already has been stopped as I can
>>not believe that removing chess knowledge is progress.
>
>Its that bloody Hyat, isn't it ? With his one paradigm, and his shouting the
>one and only way, and his driving everybody else off the ng by answering
>everything, Crafty this, Crafty that, Crafty the bloody other.
>
>So now he's convinced you THERE IS ONLY ONE WAY ?
>
>Czub - Wo bist du ?
>
>>
>>The fact that Program_X scores better against Program_Y doesn't
>>(necessarily) mean that X is better than Y. X just scores better
>>against Y and that's all. Maybe Y scores a lot better against humans
>>than X and Aegon certainly has proven that.
>>
>>So where do we stand now in computer chess?
>
>Well, I try and diversify away from from it. Too many stupid people. Maybe I
>should defeat them all and then swan off ?
Paderborn would be the right place for you. Imagine CST (4000) &&
CST (300,000) having the 100% score. The ultimate proof :)
My plea is that reviewers (journalists) when they judge the playing strength
of a chess program not only limit themselves on comp-comp and test-sets
but also emphasize on "playing style", "human-alike play", "positional
understanding", "positional sacrifices" and so on.
That would be not a bad start no?
I mean which journalist is able to tell if version 2.0 is stronger than
version 1.0? This was possible in the early days when programs
were in the 2000 area but now we are in the 2500 area so how
can a journalist judge? Can a 2200 rated journalist judge that?
When I read reviews in magazines of chess programs journalists
still do. X is stronger, Y is not stronger. Based on what?
Better divide playing strength into categories and give it a description:
. Playing style .....................
. Human-alike play .....................
. Positional understanding .....................
. Positional sacrifices .....................
. Tactics .....................
. Comp-Comp .....................
or whatever is important to know about a chess engine.
Maybe (just maybe) when chess journalists start to judge chess
programs in another way and emphasize on other points other
than comp-comp and tactics chess programmers are willing
(forced) to return to the things that really matters and that is
adding new and better chess knowledge instead of removing
chess knowledge from their programs.
Ed Schroder
Steve Maughan
>
Ed
Why not write two different programs. This is the approach that I am
taking. One will be a knowledge based approach (i.e. incremental move
generation) the other as fast as possible. In having two completely
different aims I will see which wins and hopefully evolve a better program.
Steve Maughan
Jeroen ;-}
Ed Schroder heeft geschreven in bericht >So where do we stand now in
computer chess?
>
>And more important, what is the way to go?
>
>Ed Schroder
>
>
Will Rebel "sterven in schoonheid" (sorry) or be a commercial succes?
Jeroen ;-}
Chris Whittington
Ed Schroder wrote in message <36E18390...@rebel.nl>...
Fine, but for the first time of WCCC they want money $500. And Thorsten is
banned from the icca, he's my operator, and that cancels me from sending
him, based on rule '50% of team must be icca member'. Besides, the whole
thing will be a police state designed to discover malcreant Craftys. I'm
sick of police states.
>
>My plea is that reviewers (journalists) when they judge the playing
strength
>of a chess program not only limit themselves on comp-comp and test-sets
>but also emphasize on "playing style", "human-alike play", "positional
>understanding", "positional sacrifices" and so on.
>
>That would be not a bad start no?
I've never seen a CSTal review that captures it yet. Either it's sycophantic
praise from someone who you know knows zilch about it. Or else it's
nah-nah-nah-nah, 18-2. Or else it gets ignored totally. The journalists are
the worst of the lot. They know NOTHING.
>
>I mean which journalist is able to tell if version 2.0 is stronger than
>version 1.0? This was possible in the early days when programs
>were in the 2000 area but now we are in the 2500 area so how
>can a journalist judge? Can a 2200 rated journalist judge that?
You flatter them. 2200 ? There's a handful of people in the world able to
judge a chess engine beyond anything more than SSDF results. And they all
spend their time hating each other.
>
>When I read reviews in magazines of chess programs journalists
>still do. X is stronger, Y is not stronger. Based on what?
Nonsense. That's what.
>
>Better divide playing strength into categories and give it a description:
>
>. Playing style .....................
>. Human-alike play .....................
>. Positional understanding .....................
>. Positional sacrifices .....................
>. Tactics .....................
>. Comp-Comp .....................
>
>or whatever is important to know about a chess engine.
Tell me who. Who can judge this ?
I only know Thorsten. And even he talks nonsense at times.
>
>Maybe (just maybe) when chess journalists start to judge chess
>programs in another way and emphasize on other points other
>than comp-comp and tactics chess programmers are willing
>(forced) to return to the things that really matters and that is
>adding new and better chess knowledge instead of removing
>chess knowledge from their programs.
Sure. But you give these people a lot of power. Are they up to it ?
Chris Whittington
Chris Whittington
Steve Maughan wrote in message ...
>>And more important, what is the way to go?
>>
>
>
>Ed
>
>Why not write two different programs. This is the approach that I am
>taking. One will be a knowledge based approach (i.e. incremental move
>generation) the other as fast as possible. In having two completely
>different aims I will see which wins and hopefully evolve a better program.
Or else write a two-brainer to choose which is best in any situation :))
Chris Whittington
>
>Steve Maughan
>
>
Ed Schroder
>Excuse me that I dare to enter such a typical expert question. But I
>liked your post for a very personal reason. I think that human v comp/
>comp v human playing is much more important. And we should try to
>reanimate such tournaments.
>
>You wrote about a real paradoxon. There might be a prog that wins
>against all other progs and still -- it's not necessarily the
>best/strongest 'player'.
>
>Ed, as I could see more than once, you are the guy to turn around the
>wheel if you once was convicted that the direction was wrong.
I don't think I have the power to do such a thing. I do think that "chess
journalists" is the key word here. If I (actually we) can convince them
they might change their minds and review programs differently.
>I tell you what. Forget about all the tuning over-night with autoplayer.
>Simply forget it. Have the guts to simply deny to do it. Let the rest do
>it. And I for one promise you one thing. If you can show that your
>approach leads you to better results in games vs, humans you will be the
>new leader of the pack. The actual and past tradition of imbreeding in
>comp v comp is nuts Ed, completely inhuman nonsense. And the clients
>don't want the REBEL who beats them all (machine-wise) but they want a
>hell of machine that is able to play human beings. Totally different
>thing.
I of course agree.
Just one remark, there are a lot of computer chess enthusiastic who do
love comp-comp and quick solving difficult positions and I want to serve
them too.
>Methinks that even Bob there is more advanced than you commercial guys.
>Ed, you are not forced to repeat the old tradition on and on. You even
>don't need sort of contract with your competition. Just do it. And in
>business talkie talk, _you_ are the new leader of a new trend. Period.
>
>
>All this beyond all the bad past between us. Good to see you back here.
>No joking or irony intended. I always said that you belonged into an
>open forum. :)
Agreed too :)
Ed Schroder
Rolf Tueschen
<920750549.2850.0...@news.demon.co.uk>:
>is
>banned from the icca, he's my operator, and that cancels me from sending
>him, based on rule '50% of team must be icca member'. Besides, the whole
>thing will be a police state designed to discover malcreant Craftys. I'm
>sick of police states.
The best thing to do is this, Chris. You send me as your first choice
operator. Of course I'm only there for the general politics, but I'm
_your_ Number One/UNO! The other guy, maybe, can function as the techno
brain, if you allow medical help against bites during the whole
tournament. Perhaps also for the night if operators must sleep in one
room altogether ... ;)
I see no one who could beat us. But you must assure that the guy won't
bite, or _if_ he bites that I can get immediate blood transfusions,
agreed?!!
BTW I think it's a piece of cake to give them in ICCA a call that I'm a
new member. And please ask them if I can use at least hand-cuffs during
rounds. Of course, the details -- which engine version against which
opponent is his duty. Please let me announce the decisions in public.
You know, it's important for my status as a future expert!
Any questions so far?
Chris Whittington
Rolf Tueschen wrote in message <7bs3p7$e5n$1...@news02.btx.dtag.de>...
>>banned from the icca, he's my operator, and that cancels me from sending
>>him, based on rule '50% of team must be icca member'. Besides, the whole
>>thing will be a police state designed to discover malcreant Craftys. I'm
>>sick of police states.
>
>operator. Of course I'm only there for the general politics, but I'm
>_your_ Number One/UNO!
Ok, it's a deal. Numero Uno Rolf. We will send you new version each round,
booked against Crafty, to kill it; but most important kill Friedel. He will
lose hahaha. End of the Emperium. Yes. And you will lose against Virtual, it
is $10,000, you get 10% ok, but make it realistic, you draw, then blunder
Nb8+. Easy I program it.
>The other guy, maybe, can function as the techno
>brain, if you allow medical help against bites during the whole
>tournament. Perhaps also for the night if operators must sleep in one
>room altogether ... ;)
You didn't know ? They all slept in the same hotel. It was a bath-house. All
homosexuals, Rolf, every one. You can go too, but only if you drive me to
the airport, ok ?
>
>I see no one who could beat us. But you must assure that the guy won't
>bite, or _if_ he bites that I can get immediate blood transfusions,
>agreed?!!
Right. But don't come on crutches. Bruce will veto the dispensation order
and you must move. Or else not play. If it were baseball, you wouldn't play,
no ? Nobody plays baseball on crutches, innit ?
>
>BTW I think it's a piece of cake to give them in ICCA a call that I'm a
>new member. And please ask them if I can use at least hand-cuffs during
>rounds.
Yes, Peter will be back too. He will have read the 'seven day late - there
is too much noise to signal ratio I quit' and realised it was all over. But
then he has a brain.
>Of course, the details -- which engine version against which
>opponent is his duty. Please let me announce the decisions in public.
>You know, it's important for my status as a future expert!
>
>Any questions so far?
Which drama school were you educated at ? They did good supporting actor
design, no ?
Chris Whittington
>
>
>
Rolf Tueschen
>>From: TUESCHEN.MEDIZ...@t-online.de (Rolf Tueschen)
>>Ed, as I could see more than once, you are the guy to turn around the
>>wheel if you once was convicted that the direction was wrong.
>I don't think I have the power to do such a thing. I do think that "chess
>journalists" is the key word here. If I (actually we) can convince them
>they might change their minds and review programs differently.
Ed, it's a matter of perspective. From the outside, believe me for one
time, it looks different. Of course you set the standards as
programmers/business guys. Just one idea. Do you think that "chess
journalists" would make difficulties. Didn't you mean "computer chess
journalists"? I think Chris said what had to be said. Where do you see
for instance one single 2200 cc journalist? Prove me ignorant!
>>I tell you what. Forget about all the tuning over-night with autoplayer.
>>Simply forget it. Have the guts to simply deny to do it. Let the rest do
>>it. And I for one promise you one thing. If you can show that your
>>approach leads you to better results in games vs, humans you will be the
>>new leader of the pack. The actual and past tradition of imbreeding in
>>comp v comp is nuts Ed, completely inhuman nonsense. And the clients
>>don't want the REBEL who beats them all (machine-wise) but they want a
>>hell of machine that is able to play human beings. Totally different
>>thing.
>I of course agree.
>Just one remark, there are a lot of computer chess enthusiastic who do
>love comp-comp and quick solving difficult positions and I want to serve
>them too.
Ed, do you want to make me immortal? :)
I give you here as the only one the following idea under usual
copyright. But you can use it if you want. Perhaps you know the
ChessBase/Fritz/Mate engine concept? Just try to make an engine version
of REBEL that does exactly what you mentioned. For the sake of fun,
write a naked-hell-as-fast engine for autoplayers. Then write a version,
just call it _R o l f i n g m o d e_ [with a tongue in cheek] where
REBEL does search especially in complicated tactical positions as fast
as hell again. Look at the 'Fritz correspondance analysis mode'. Just to
understand my idea of several modes. The content is different of course.
But to come back to our idea, your main playing REBEL should be the one
tuned against human beings. I cannot tell you how happy you made me with
your post.
Now to your justified concern about those who like to analyse comp v
comp. I wouldn't defame such a hobby. And in the beginning of all the cc
it was the only thing you could do because comps/micros simply were too
weak for in detail human play. But we have a different situation now. In
the end most of the guys will understand that now it's time to
reconsider what they are doing. But as I said above I would especially
provide this group with an autoplayer version. Someone asked here if you
wanted to die in beauty. I think you were the first who simply
understood the new times. We endusers want a strong "human" machine. BTW
I have some ideas about the books. But I shouldn't talk here in public.
Why giving away all new ideas?
(Ed, I hope you get a first impression how stupid any exclusion even of
the most stupid newbie might be, a fallacy so to speak. Human beings
have very different strengths. What you techno beasts need is the help
of people who are _not_ in the business but close enough to know what cc
program is all about _and_ able to do so-called 'brainstorming' to
provide you with "new" or notr so new ideas. Know what I mean? Let's
start that. Life is too short ... I'm waiting with literally minimum 30
new ideas a tempo. Now think of all the other members too. Do you really
believe that burnt out older experts will have all the new ideas
themselves?)
Rolf Tueschen
>Which drama school were you educated at ? They did good supporting actor
>design, no ?
You misunderstood me. I wasn't joking. I was just a matter of, err, I
simply learnt of NOT telling the truth in case it's of TITANIC
extensions. You simply can't do that I was told. Therefore the actual
hype. But it _was_ real!
>Chris Whittington
Robert Hyatt
:>From: TUESCHEN.MEDIZ...@t-online.de (Rolf Tueschen)
:>Excuse me that I dare to enter such a typical expert question. But I
:>liked your post for a very personal reason. I think that human v comp/
:>comp v human playing is much more important. And we should try to
:>reanimate such tournaments.
:>
:>You wrote about a real paradoxon. There might be a prog that wins
:>against all other progs and still -- it's not necessarily the
:>best/strongest 'player'.
:>
:>Ed, as I could see more than once, you are the guy to turn around the
:>wheel if you once was convicted that the direction was wrong.
: I don't think I have the power to do such a thing. I do think that "chess
: journalists" is the key word here. If I (actually we) can convince them
: they might change their minds and review programs differently.
:>I tell you what. Forget about all the tuning over-night with autoplayer.
:>Simply forget it. Have the guts to simply deny to do it. Let the rest do
:>it. And I for one promise you one thing. If you can show that your
:>approach leads you to better results in games vs, humans you will be the
:>new leader of the pack. The actual and past tradition of imbreeding in
:>comp v comp is nuts Ed, completely inhuman nonsense. And the clients
:>don't want the REBEL who beats them all (machine-wise) but they want a
:>hell of machine that is able to play human beings. Totally different
:>thing.
: I of course agree.
: Just one remark, there are a lot of computer chess enthusiastic who do
: love comp-comp and quick solving difficult positions and I want to serve
: them too.
:>Methinks that even Bob there is more advanced than you commercial guys.
:>Ed, you are not forced to repeat the old tradition on and on. You even
:>don't need sort of contract with your competition. Just do it. And in
:>business talkie talk, _you_ are the new leader of a new trend. Period.
:>
:>
:>All this beyond all the bad past between us. Good to see you back here.
:>No joking or irony intended. I always said that you belonged into an
:>open forum. :)
: Agreed too :)
: Ed Schroder
Just remember this: In every 'world' there are lots of metrics. In
automobiles, we have 0-60 acceleration times, 60-0 braking times,
skid-pad lateral acceleration (G-forces), slalom times, top speed,
quietest, most comfortable, etc. IE there is _no_ one automobile that
will rank #1 in every category, because they are _mutually exclusive_
in many cases.
Trying to write a chess program to beat other programs, beat anti-
computer players, beat GM's, be the fastest tactical solver around,
and so forth is probably unrealistic. You might do one, or maybe even
two of those. But not _all_ unless you throw the resources of something
like "Deep Blue" at the problem.. Then you can beat all other programs,
solve everything impossibly fast, and also play with the top GM players.
But when money is a consideration, goals have to be scaled back. For me,
I do what I consider to be 'fun'... namely trying to take the scalps of
GM players.
Albert Silver
Chris Whittington wrote:
> Ed Schroder wrote in message <36E128CC...@rebel.nl>...
>
> Er ...... this is *the* Ed Schroder talking to me ?
>
> Yup, checked the header file. It is he.
>
> Why this ? From the one who played such a large part in the unwarranted
> censorship of my goodself for all that time ?
>
> Anyway, polite as I am, I shall endeavour to reply to your post ....
>
>
> < snip>
>
> The bad news is that any drongo can do this. Worse, any drongo does do this.
> Once I used to think chess programmers had something special. Therefore
> their minds were high quality. Unfortunately, listening to them on the news
> groups, with words, ideas, stuff that provokes thought ... there is none.
> Chess programmers are dorks. Simple but true. If they had brains, they'ld
> show it with words, understanding and ideas. Only they don't. Look to the
> top and what do you see ? An idiot.
Feeling better now that you got it off your chest?
>
>
> So what we got ? A whole load of drongos who easily as hell can put
> bean-counters together, and these bean-counters can play. Play shit style
> but play.
>
> >
> >As a result (also because of today's fast Pc's) you get your program
> >on incredible deep ply-depths. 13 and 14 plies in the middle game are
> >no exceptions anymore.
>
> Right. My first bean-counter (I just described it), I had it playing 5 ply,
> full width, piece square tables only, no pruning. Came back with move
> immediate. Just to see if it worked. All connetced to the CSTal WIn 2
> interface. The bloody thing smashed me.
>
> CSTal2 doesn't smash me, it plays positional, I can understand the game,
> there are themes, it's consistent. Probably Rebel is like this for you too.
> If I lose, its because I'm ground down, I can see why and how. But this
> 5-ply thing, I had no respect for, I just played my usual positional stuff,
> not really thinking, and it random killed me. Not because it played for it,
> not because of a theme, not for any understandable chess reason. Just a
> random minefield. Oh, whoops, I trod on another mine.
>
> This is not chess, it's just stupid pointless nonsense. Might as well toss a
> coin or play Russian roulette.
>
> This is how it is a 5 ply, with me not bothering to think. It's the same
> random minefield at 15 ply. Only the fools don't realise. They don't realise
> because they can't play chess. Can't play chess, buy a bean-counter -
> simple. And they do.
You sound bitter there. I don't think it's that simple. I don't think they buy
bean-counters, as you put it, because they can't play chess, and the way you
imply it, don't care. True, a weaker player by the very description does not
play a great game of chess, and therefore, when confronted with the choice of
available programs, cannot simply play them and see which one is stronger or
better suited. The player will have to read articles, listen to the experts or
at least people he or she believes to be in the know. Yet it is the experts who
are declaring so and so program stronger as it outcalculated, or as you put it,
'outmined' the other program. Not the weaker players buying them. Blame the
experts, blame the media. Not the consumers.
>
> >
> >This subject has been my attention since a couple of months now.
> >It's not that simple as you have for instance Hiarcs and Chessmaster
> >which are "slow searchers" (a very low NPS) and they both are doing
> >fine in comp-comp.
>
> >However if you take a closer look both programs seems to be
> >specialized in tactics unless the fact they have a low NPS.
Note to Schroeder: Hiarcs 'specialized' in tactics? Although it may excel at
tactics as well, I think Hiarcs has a huge edge over other programs in it's
endgame play.
>
> >
> >The last months I have worked on Rebel adding tactics, more pruning,
> >a lower branch factor. As a result it performs better in comp-comp but
> >I definitely see a drop in the positional play. The quality of the
> >returned
> >moves is simply less, still it is a better comp-comp fighter now.
> >
> >What to do?
>
> Dunno ? Do you care that much ?
>
> I know I can make something that will wipe them out. Really now a question
> of whether I can be bothered to do it.
:-) Just dying to have someone say it aren't you? Of course, it may not be quite
as interesting coming from a non-chess-programmer (though maybe not, considering
your opinion of them), but I'll say it:
I dare you!
<snip>
Albert Silver
mclane
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
This is what I have always said. but i am sure, nobody will believe
you ed.
if x scores better against y, than x shall be stronger in the paradigm
of a materialist, having statistics and elo-rating-lists :-)))
>So where do we stand now in computer chess?
>And more important, what is the way to go?
>Ed Schroder
The fight you describe is the fight many more intelligent programs had
to compete over years against the fast-searchers.
if i take e.g. rebel10c and hiarcs7.01 i am sure
hiarcs is the strongest, followed by rebel.
(i have no experience so far with the tiger-option in rebel).
than other programs follow.
so in fact at least 2 more intelligent programs compete over
fast-searchers.
in the years before, it was mchess dominating, also a slow program.
as you know, chris always tried the more etxreme way by trying to
catch the fast-searchers with an engine that makes only 4500 NPS.
so there have been efforts.
cstal was not unsuccesfully in paris.
so the distance between fast-searchers and intelligent programs is not
that big.
what is the way to go.
i don't believe that fast-searchers with primitive knowledge will be
the way.
would you buy a fast but unconvinient car ?
Of course some people would. but the mass of people
want to drive convinient. and fast.
would you marry a stupid girl ?
I am sure some people do. they fell confident with the fact that she
looks nice, but after 20 years ? what do you want to talk with a
stupid girl over 20 years ?
no -
i hope the slow-searchers do not die out.
i like to drive citroen. not porsche.
best wishes
mclane
>Yup, yup, yup. Actually the conclusion is that *Berliner's* encoded
>knowledge (HH minus HL) was worth 1 ply. I think he wanted to work up the
>argument that Botwinnik's knowledge was therefore also only worth one ply.
right. BERLINER's knoweldge was worth one ply. this does not need to
fit to the knowledge other people implement in programs. or ?
>Note how nobody ever questioned the knowledge. it came from the great man,
>so it must have been *the_knowledge*, innit ?
! :-))) exactly.
>Anyway, such is the power of the academic research nonsense that it goes
>down in history for bean-counter justification regurgification whenever
>necessary.
! :-))) bean counter found bean counter myths...
>Both you and I know that (value of knowledge > 1 ply). Depends on the
>knowledge and the position.
>Yup. yup, yup.
>Boring isn't it ?
>And it's so easy. Which is why there are so many programs. I work at the
>moment with an assistant programmer. He can play chess, which helps. In the
>old days, I used to think chess programming was a huge complex process, but
>that was because I used to do everything, user interface, QA, bug testing
>blah-blah-blah. Now, with second programmer, I zap out the code, and he (and
>me too if necessary) clean it up and make sure it works. This way, I keep
>the overview, and avoid getting bogged down in the detail.
! I hope you are right. :-)))
>It took us two months to write the nitty-grittys of a bean-counter. That was
>the start point of the new project, then to go onto higher things. Based on
>better than bitmaps. Similar idea, but better.
>Then, to go bean-counter, just a bit of eval, and the usual tricks. Bob is
>your uncle.
>The bad news is that any drongo can do this. Worse, any drongo does do this.
>Once I used to think chess programmers had something special. Therefore
>their minds were high quality. Unfortunately, listening to them on the news
>groups, with words, ideas, stuff that provokes thought ... there is none.
>Chess programmers are dorks. Simple but true. If they had brains, they'ld
>show it with words, understanding and ideas. Only they don't. Look to the
>top and what do you see ? An idiot.
>So what we got ? A whole load of drongos who easily as hell can put
>bean-counters together, and these bean-counters can play. Play shit style
>but play.
:-))))
thats the way the world works.
nobody has ever said windows is a great product.
but it dominates.
>Right. My first bean-counter (I just described it), I had it playing 5 ply,
>full width, piece square tables only, no pruning. Came back with move
>immediate. Just to see if it worked. All connetced to the CSTal WIn 2
>interface. The bloody thing smashed me.
! :-)))
>CSTal2 doesn't smash me, it plays positional, I can understand the game,
>there are themes, it's consistent. Probably Rebel is like this for you too.
>If I lose, its because I'm ground down, I can see why and how. But this
>5-ply thing, I had no respect for, I just played my usual positional stuff,
>not really thinking, and it random killed me. Not because it played for it,
>not because of a theme, not for any understandable chess reason. Just a
>random minefield. Oh, whoops, I trod on another mine.
! :-)))
>This is not chess, it's just stupid pointless nonsense. Might as well toss a
>coin or play Russian roulette.
this is not chess, it is the emulation of it.
right ! exactly my opinion. they win, but cannot play chess.
and they argue that they play "better" because they win.
they are idiots.
>This is how it is a 5 ply, with me not bothering to think. It's the same
>random minefield at 15 ply. Only the fools don't realise. They don't realise
>because they can't play chess. Can't play chess, buy a bean-counter -
>simple. And they do.
right. they buy fritz5 , and they get what they deserve.
made from idiots, for idiots :-)))
>>What if a program with a lot of chess knowledge (having a low
>>NPS because of that) is faced with such a "fast searcher" AND the
>>HH_HL principle is true. It will lose no? The "fast searcher" naturally
>>comes 3-4 plies deeper because of a smart search and the HH_HL
>>principle ensures victory after victory against a low NPS program.
>>
>>True sofar?
>No.
>What is true is, what is the point of trying to make knowledge program that
>can play with style, themes, and chessisms, when the masses don't have the
>brainpower to understand it, and hordes of dork programmers can knock up
>competing bean-counters in a few months ?
right. thats a problem. but - we will stand this, or ?
>Right. In the land of the blind, having a longer feely stick works wonders.
>Right. Bastards.
>>- improved selected search (smart pruning) to get a low branch factor;
>>- make the program more aggressive to ensure tactics in comp-comp games;
>>
>>As a result the programs do a lot better now in the comp-comp area.
>Yup. Boring isn't it ?
>>Is this the end of adding new chess knowledge to our programs?
>Nah. You just got to get the dynamic knowledge encoded. How do you think
>CSTal copes at 4 knps against 300knps ? Or how about CSTal at 4knps to get
>the good positions, then switch to bean-counter CSTal at 300 knps to finish
>them off ?
right. i think you have partly showed that it works, chris.
>It gets a lot of win positions that it then can't convert right now.
>The dorks are frightened, believe me.
you can change this. use a stupid engine for winning the won games.
>Sure. If nobody can see anything, then may as well cut away every redundant
>thing, and rely on the feely stick.
>Only needs one man with an eye and they are all kaput.
>Dunno ? Do you care that much ?
>I know I can make something that will wipe them out. Really now a question
>of whether I can be bothered to do it.
i hope you do it.
>Stupid. Stupid. Too many stupids. And they have a tendency to lie as well.
! :-)))
they live from doing it that way !
>Shall we jump ?
No !
>Its that bloody Hyat, isn't it ? With his one paradigm, and his shouting the
>one and only way, and his driving everybody else off the ng by answering
>everything, Crafty this, Crafty that, Crafty the bloody other.
>So now he's convinced you THERE IS ONLY ONE WAY ?
>Czub - Wo bist du ?
i don't think we need to convince ed.
he is still on "our" side.
As far as I see, ed has made friendship to christophe theron, to
handle the comp-comp stuff better, so maybe he can further concentrate
on making rebel more intelligent.
friendship is one way to stand against the fast searchers.
we need to convince ed. he is not convinced, he seems to be
frustrated. that is not convinced.
>Well, I try and diversify away from from it. Too many stupid people. Maybe I
>should defeat them all and then swan off ?
>Upwards, Ed.
>And away from Hyatt. And the Great Satan.
>Chris Whittington
?? :-)))
best wishes
mclane
>This applies to Chessmaster, but not to Hiarcs 7, and it is an important
>exception to the general rule you mention. Hiarcs is by far the slowest NPS
>and does great in comp-comp games at any speed, including blitz. I tried H7
>lowering its selectivity from 5 (default) to 2, making it less tactical and
>more knowledge based, and it still does great. Is this enough to dismantle
>your theory?
hiarcs 7 is better in tactics than hiarcs6.
so what ?
>As for the drop of quality in positional play, it is not obvious to me
>either. Aside from the already mentioned Hiarcs 7, Junior 5 and Tiger play a
>quality game that I never saw before in fast searchers, and very rare in slow
>searchers even today.
i have to disagree.
junior5 plays ugly and passive, with senseless main_lines.
best wishes
mclane
>>Dunno ? Do you care that much ?
>Yes I do.
>I care because the playing strength of programs is judged by playing
>against other programs also by running test-sets comparing solution
>times of key moves.
because the magazins that write the articles are a bunch of idiots...
nobody of them takes enough time to value the playing strength.
all they do is engine-engine and present the data.
idiots.
>So in order to get a positive review your program need to perform
>well in comp-comp and in these test-sets.
right. but these magazins are leaded by idiots.
I do understand that you have a problem ed,
you compete in the market. chris is not competing in this market.
he does not to make money. he can do whatever he wants.
and he does.
therefore i am looking forward in you making team-work with
good programmers. you can exchange the best of your programs, and
maybe
find a new way.
>I think your idea to combine CST (4000) and CST (300,000) sounds
>convincing enough to give it a try.
sound much like rebel + tiger ?? :-)))
>Mayday, mayday.... still wet :)
:-)))
>Paderborn would be the right place for you. Imagine CST (4000) &&
>CST (300,000) having the 100% score. The ultimate proof :)
! :-)))
>My plea is that reviewers (journalists) when they judge the playing strength
>of a chess program not only limit themselves on comp-comp and test-sets
>but also emphasize on "playing style", "human-alike play", "positional
>understanding", "positional sacrifices" and so on.
If they magazine allows the testers to think. and if the testers are
capable to think.
>That would be not a bad start no?
this should be the normal standard. but - we all know - it isn't.
these guys there concentrate 4 hours on a sunday before dead-line, let
the program autoplay engine-engine on blitz-level, and right that the
program is good, but not that good as fritz is. then the whole thing
is written in chess-base-magazine CSS and some reader-letters are
published who praise fritz, kill mcs.
this is not the end of the universe.
this is the beginning.
i don't think these shallow people are the key to the world.
they are the way to mc_donalds, but not the key to the world.
i do also believe that, if you want tell the people about the
differences between fast-searchers, and the different things both
concepts can give, they would not always choose fritz5.
but it is the method of those chessbase dominated magazines
(css, schach64, ...) that they only praise chessBase products, in a
more or less sufficient way bill gates takes care that windows wins
anything.
>. Playing style .....................
>. Human-alike play .....................
>. Positional understanding .....................
>. Positional sacrifices .....................
>. Tactics .....................
>. Comp-Comp .....................
they have tried in css, you know what happened to hiarcs ??
best wishes
Stuart Cracraft
>As a result (also because of today's fast Pc's) you get your program
>on incredible deep ply-depths. 13 and 14 plies in the middle game are
>no exceptions anymore.
>
>What if a program with a lot of chess knowledge (having a low
>NPS because of that) is faced with such a "fast searcher" AND the
>HH_HL principle is true. It will lose no? The "fast searcher" naturally
>comes 3-4 plies deeper because of a smart search and the HH_HL
>principle ensures victory after victory against a low NPS program.
>
>True sofar?
>
>What do you think of the following statement of mine:
>
>In comp-comp a smart search is worth 200-300 ELO points.
>
>IMO many programmers have done the following the last years:
>- remove (!!) chess knowledge from the program to get a high NPS;
>- improved selected search (smart pruning) to get a low branch factor;
>- make the program more aggressive to ensure tactics in comp-comp games;
>
>As a result the programs do a lot better now in the comp-comp area.
>
>Is this the end of adding new chess knowledge to our programs?
>
>It bothers me. To compete in the comp-comp area you have to
>remove chess knowledge???
>
>The bad news is, it apparently seems to work!
>
>This subject has been my attention since a couple of months now.
>It's not that simple as you have for instance Hiarcs and Chessmaster
>which are "slow searchers" (a very low NPS) and they both are doing
>fine in comp-comp.
>
>However if you take a closer look both programs seems to be
>specialized in tactics unless the fact they have a low NPS.
>
>The last months I have worked on Rebel adding tactics, more pruning,
>a lower branch factor. As a result it performs better in comp-comp but
>I definitely see a drop in the positional play. The quality of the
>returned
>moves is simply less, still it is a better comp-comp fighter now.
>
>What to do?
>
>Every time a new program is released the program is judged in
>magazines (or internet) because it is compared with other chess
>programs and the score it gained against them.
>
>To compete in this area you have to make your program a weaker
>positional player????
>
>And the fact that important human-comp tournaments (AEGON,
>Harvard cup) are gone makes comp-comp even more important
>or even worse you can ask yourself the question: what else is left?
>
>Thinking all of this over the last months I came more or less to the
>conclusion that we have entered a new area, that adding new extra
>chess knowledge works out counter productive and that progress
>in computer chess will stop or already has been stopped as I can
>not believe that removing chess knowledge is progress.
>
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
>
>So where do we stand now in computer chess?
>
>And more important, what is the way to go?
>
>Ed Schroder
>
>
The right thing to do is not to depend on computer-vs-computer ratings
or results entirely for the computer chess program.
Have perhaps 25-50% of a program's rank based on that.
Have another 50-75% of the same program's rank depend upon its play
on FICS.
Sure, people there will use computers and ratings will be artificially
lower; but it won't be this awful skew from computer/computer playing.
--Stuart
Dann Corbit
There is too much focus on tactics. He is also concerned that the SSDF may
drive sales, to the detriment of a program's ability to play against humans.
Again a valid point.
I think we see an internal annoyance. Should a programmer aim for the
market, or for what his/her heart says is the best chess program? I think
that Vincent Diepeveen epitomizes the "Add knowledge to the engine and piss
on tactics" approach. Certainly others come to mind as well. I would like
to get a copy of CS-Tal, but I can't find any chess program vendors that
sell it nearby. Just from tracing over some games, I see a very exciting
brand of chess.
I don't think that you should sell Hiarcs short, though. Hiarcs is a
wonderful program with many advanced features. And it does not play boring
chess (at least not in my view). In fact, if you look at the SSDF games,
they are littered with 120 move draws. Talk about a yawner. But Hiarcs has
a lot of scintillating slugfests with other programs.
Now, what about market realities? I think programmers face a dilemma here.
Shall they write programs that can win SSDF games or programs that play
better strategic chess? This is not a simple question with an easy answer.
When your programs put bread on your table, to some degree you must bend to
serve what people want, even if what they want is stupid. On the other
hand, if any real breakthrough is made in chess programming it will come
from positional and strategic understanding, not from one more ply of
tactics.
If you write advanced programs that play interesting chess, you will always
have some market. I think that better advertising and a web presence might
be helpful to the smaller chess shops. But then, you have to have the money
to spend on advertising in the first place. Perhaps a partnership of some
sort with a large software house might be in order.
I think that improved annotation and better database capabilities are
important missing features of current chess programs. Perhaps if some chess
program companies partnered with the chess database companies they could
produce some truly remarkable programs. But then again, what I want may be
far-flung from what the broad audience is interested in.
The improvement of chess programs is not going to stop. There are too many
people interested in doing just that. If the best and brightest minds leave
to pursue something else, then it may be a setback. But eventually, someone
will pick up the baton and run with it. There will also be more
breakthroughs. Perhaps IBM will refine and market their chess chip that was
used in Deep Blue. If mass produced, it would not need to be much more
expensive than any other CPU. Perhaps some revolution will occur that
fundamentally increases our understanding of the game of chess.
I don't think we're standing on the edge of a cliff. We're at the base of
the mountain and the climb is only just beginning.
--
C-FAQ: http://www.eskimo.com/~scs/C-faq/top.html
"The C-FAQ Book" ISBN 0-201-84519-9
C.A.P. Newsgroup http://www.dejanews.com/~c_a_p
C.A.P. FAQ: ftp://38.168.214.175/pub/Chess%20Analysis%20Project%20FAQ.htm
--
C-FAQ: http://www.eskimo.com/~scs/C-faq/top.html
"The C-FAQ Book" ISBN 0-201-84519-9
C.A.P. Newsgroup http://www.dejanews.com/~c_a_p
C.A.P. FAQ: ftp://38.168.214.175/pub/Chess%20Analysis%20Project%20FAQ.htm
Greg Kennedy
>- remove (!!) chess knowledge from the program to get a high NPS;
>- improved selected search (smart pruning) to get a low branch factor;
>- make the program more aggressive to ensure tactics in comp-comp games;
>As a result the programs do a lot better now in the comp-comp area.
>Is this the end of adding new chess knowledge to our programs?
>It bothers me. To compete in the comp-comp area you have to
>remove chess knowledge???
>The bad news is, it apparently seems to work!
>The last months I have worked on Rebel adding tactics, more pruning,
>a lower branch factor. As a result it performs better in comp-comp but
>I definitely see a drop in the positional play. The quality of the returned
>moves is simply less, still it is a better comp-comp fighter now.
>What to do?
>Every time a new program is released the program is judged in
>magazines (or internet) because it is compared with other chess
>programs and the score it gained against them.
>To compete in this area you have to make your program a weaker
>positional player????
>So where do we stand now in computer chess?
>And more important, what is the way to go?
For commercial chess programmers, there is a
solution: set the default mode of your program
for computer/computer success, with high nodes
per second and minimal chess knowledge.
Under your options menu, place a setting which
changes to your other mode- anti-human, with
tons of chess knowledge and low nodes per
second.
This nets your program high ratings vs. the
other computer programs, without giving up the
option of better knowledge. More work for you,
but I have seen that many programs already have
different settings which entailed more work for
the programmers. The user can then decide
what they want- results vs. computers, or more
knowledge. They buy your program, and get
BOTH.
- Greg Kennedy
Phil Innes
To compete in this area you have to make your program a weaker
positional player????
And the fact that important human-comp tournaments (AEGON,
Harvard cup) are gone makes comp-comp even more important
or even worse you can ask yourself the question: what else is left?
Human comp tournaments played by normal chess rules would
always result in the defeat of the computer by humans rated
2000. Without the opening book, is any program stronger than
1900, even running at 500Mhz?
Thinking all of this over the last months I came more or less to the
conclusion that we have entered a new area, that adding new extra
chess knowledge works out counter productive and that progress
in computer chess will stop or already has been stopped as I can
not believe that removing chess knowledge is progress.
The fact that Program_X scores better against Program_Y doesn't
(necessarily) mean that X is better than Y. X just scores better
against Y and that's all. Maybe Y scores a lot better against humans
than X and Aegon certainly has proven that.
So where do we stand now in computer chess?
What do people use their programs for? Storing games, an
on-line interface, to learn openings, to test tactical skills.
All this seems constant.
Some time ago Chris down-played my comment about the opening
book, but Bob seemed to agree with me. I don't understand
Chris' objection.
Ed is very honest in pointing out the basis of current
development and improvements. But these seem like very small
increments to me.
And more important, what is the way to go?
Learn why humans play (potentially) better chess. Emulate it.
Get off your "we the programmers control the thing entirely"
attitude and revisit the subject by inviting other opinion.
The crux of the thing is, no matter how fast the linear
processor is - all positional values are pre-ordained (before
move 1) by the programmer. These values do not change during
the game (although value set b, then, set c may be invoked)
and neither does the program learn anything.
(This has also been a problem in a broader field of robotics.
Advances were made by interacting with pyschologist, then with
choleopterists.)
Phil Innes
Ed Schroder
Ed Schroder
Thanks that so many people reacted, I don't know were to start in
replying. Sofar I have only seen postings who more or less agree
with the very basic problem I have tried to explain. I would love to
see the opinion of people who have opposite feelings, this for the
value of the discussion.
I can imagine people say, "Ed shut up, comp-comp is *the* way
to measure progress and playing strength", and defend it.
Anyone?
>> However if you take a closer look both programs seems to be
>> specialized in tactics unless the fact they have a low NPS.
>
>This applies to Chessmaster, but not to Hiarcs 7, and it is an important
>exception to the general rule you mention. Hiarcs is by far the slowest NPS
>and does great in comp-comp games at any speed, including blitz. I tried H7
>lowering its selectivity from 5 (default) to 2, making it less tactical and
>more knowledge based, and it still does great. Is this enough to dismantle
>your theory?
I am not so sure. Chessmaster and Hiarcs7 are close in (low) NPS.
The strongest point of "de Koning" programs has always been tactics
and I consider it as the best around since the introduction of the Chess
Machine (1991-1992)
I have always said Chessmaster is under-rated on SSDF. I have read
SSDF is now testing CM6000. Take my word for it, it will end very high in
the list if proper testing is done and the CM opening book (for a change)
is a good one. It belongs in the top in the comp-comp area and it wouldn't
surprise me at all if it will top. The key word of CM is tactics and it
guarantees a good comp-comp result.
About Hiarcs7, it is my opinion that besides being a good positional
player Hiarcs7 is now a tactical monster too. It seems to find everything,
CM alike.
>It is true that programs that do best in comp-comp games are very aggressive,
>to the point that their performance in my tournament was directly
>proportional to aggressiveness. But it is wrong to say that this is important
> “to ensure tactics in comp-comp games”. If this were the case, Nimzo 99, as
>much of a fast finder as anything, would take advantage from the
>aggressiveness of the opponents, and it doesn’t because it is rather passive
>and the tactical positions are built against it. The best programs take
>advantage of their aggressiveness by building up positions that favor them,
>very much like human grandmasters do. There is no such a thing as a passive
>GM. They may be positional, quiet, a la Karpov, but never passive, and some
>programs are and pay for it. Not knowing enough to play quiet and long range
>plans, the best programs play aggressive and create positions that give them
>more chances than to the opponent. You can consider this characteristic as
>pre-tactical, as an “intelligent” way to make them play, applied today by
>some slow and fast searchers. So you could say that being aggressive is
>important “to ensure tactics in comp-comp games after creating a favorable
>position”, which is a manifestation of knowledge.
>
>I read statements made by Donninger and Théron claiming that they wanted to
>get rid of “unnecessary” knowledge, and even not being a programmer it makes
>an awful lot of sense to me. It is clear that knowledge has a different role
>now at 400K N/S than 16 years ago when Mephisto III ran at 2 nodes per second
>and the poor thing had to compensate for it. Maybe we are not talking about
>further development of knowledge, but about a redefinition of what kind of
>knowledge matters more than the search being slowed down by it. Alright, I
>already said I am no programmer, but this is how things look like to me from
>the "“outside"”
>
>As for the drop of quality in positional play, it is not obvious to me
>either. Aside from the already mentioned Hiarcs 7, Junior 5 and Tiger play a
>quality game that I never saw before in fast searchers, and very rare in slow
>searchers even today.
Keyword: put your pieces as active as possible, tactics come by itself
then. Not much chess knowledge (code) is needed and it ensures a
high NPS.
>By the way, the lowering of the branching factor doesn’t necessarily apply
>either as a common trend in new engines. For instance, Junior 5, in my opinion
>one of the two very best engines around, has the lowest of all together with
>Genius, even lower than Hiarcs and much lower than Rebel.
Note that you can add tactics in several ways, a low branch factor is just
one of the possibilities. An other possibility is to add a lot of extensions with
a high branch factor as result. Deep Blue is perhaps the best example. But
the key-word remains tactics.
Ed Schroder
>Enrique
Ed Schroder
>Ed
>
>Why not write two different programs. This is the approach that I am
>taking. One will be a knowledge based approach (i.e. incremental move
>generation) the other as fast as possible. In having two completely
>different aims I will see which wins and hopefully evolve a better program.
>
>Steve Maughan
In a way this approach has already been started when I joined forces with
Christophe (Chess Tiger). We more or less agreed that for the near future
I will focus on "human-comp" and make Rebel an attractive as possible
chess engine for humans where Christophe will concentrate on comp-comp
only.
It's my opinion there lots of people who love comp-comp over human-comp,
the only thing I don't know is the division in percentage of these 2 groups.
Ed Schroder
Ed Schroder
>because the magazins that write the articles are a bunch of idiots...
>nobody of them takes enough time to value the playing strength.
>all they do is engine-engine and present the data.
>idiots.
comp-comp is just one aspect which currently is given a too high
attention value. Maybe we just should convince the computer chess
press there is more under the sun than just comp-comp?
I would say we just give them a chance.
Since Rebel6/7 (the 486/P90 days) I am wondering *how* the chess
press is able to say something about improvement of chess strength.
Most of them are in the range of 1800-2200, so how can they judge?
In my opinion most of them can not and in my opinion most of them
know it too. An escape is found to play comp-comp matches as it
gives fast results and then the final judgement comes, stronger cq
not stronger.
Having read hundreds of program reviews through the years I have
seen the most horrible mistakes because of that. I am not speaking
of myself, I mostly got away with a good review.
The above is not meant as an attack on the chess press that they
deliberately have given false information. This is not what I mean to
say. Going back to the 486/P90 days I noticed a new tendency, because
of the increasing playing strength (due to hardware and software) most
(not all) chess journalist are not able anymore to judge, as it is above
their understanding what's going on. They only fail to admit that in their
reviews. I can understand this, readers don't want to hear "I don't know"
but it is my opinion this is true for many.
I don't envy the chess press having the obligation to tell their readers
what they like yo hear, "is it stronger?"......... I would say mission
impossible these days for the most. I have difficulties myself to
estimate the progress of Rebel every time while I know every little
corner of the program.
>>So in order to get a positive review your program need to perform
>>well in comp-comp and in these test-sets.
>right. but these magazins are leaded by idiots.
No idiots, but a lack of sufficient elo maybe?
>. Playing style .....................
>. Human-alike play .....................
>. Positional understanding .....................
>. Positional sacrifices .....................
>. Tactics .....................
>. Comp-Comp .....................
>they have tried in css, you know what happened to hiarcs ??
I think the approach (is) was good, the contents were horrible I agree.
Ed Schroder
Ed Schroder
>>I don't think I have the power to do such a thing. I do think that "chess
>>journalists" is the key word here. If I (actually we) can convince them
>>they might change their minds and review programs differently.
>
>Ed, it's a matter of perspective. From the outside, believe me for one
>time, it looks different. Of course you set the standards as
>programmers/business guys. Just one idea. Do you think that "chess
>journalists" would make difficulties. Didn't you mean "computer chess
>journalists"? I think Chris said what had to be said. Where do you see
>for instance one single 2200 cc journalist? Prove me ignorant!
Enrique is. As a result I see very well balanced wordings from his side
when the subject is playing strength. Like to add Enrique is a comp-comp
lover still he manages to write good reviews about playing strength so it
can be done.
>>>I tell you what. Forget about all the tuning over-night with autoplayer.
>>>Simply forget it. Have the guts to simply deny to do it. Let the rest do
>>>it. And I for one promise you one thing. If you can show that your
>>>approach leads you to better results in games vs, humans you will be the
>>>new leader of the pack. The actual and past tradition of imbreeding in
>>>comp v comp is nuts Ed, completely inhuman nonsense. And the clients
>>>don't want the REBEL who beats them all (machine-wise) but they want a
>>>hell of machine that is able to play human beings. Totally different
>>>thing.
>
>>I of course agree.
>
>>Just one remark, there are a lot of computer chess enthusiastic who do
>>love comp-comp and quick solving difficult positions and I want to serve
>>them too.
>
>Ed, do you want to make me immortal? :)
>
>I give you here as the only one the following idea under usual
>copyright. But you can use it if you want. Perhaps you know the
>ChessBase/Fritz/Mate engine concept? Just try to make an engine version
>of REBEL that does exactly what you mentioned. For the sake of fun,
>write a naked-hell-as-fast engine for autoplayers. Then write a version,
>just call it _R o l f i n g m o d e_ [with a tongue in cheek] where
>REBEL does search especially in complicated tactical positions as fast
>as hell again. Look at the 'Fritz correspondance analysis mode'. Just to
>understand my idea of several modes. The content is different of course.
I leave that to Christophe.
>But to come back to our idea, your main playing REBEL should be the one
>tuned against human beings. I cannot tell you how happy you made me with
>your post.
>
>Now to your justified concern about those who like to analyse comp v
>comp. I wouldn't defame such a hobby. And in the beginning of all the cc
>it was the only thing you could do because comps/micros simply were too
>weak for in detail human play. But we have a different situation now. In
>the end most of the guys will understand that now it's time to
>reconsider what they are doing. But as I said above I would especially
>provide this group with an autoplayer version. Someone asked here if you
>wanted to die in beauty. I think you were the first who simply
>understood the new times. We endusers want a strong "human" machine. BTW
>I have some ideas about the books. But I shouldn't talk here in public.
>Why giving away all new ideas?
Again I entirely agree with you. Like to add that Chris was the first one
who described the problem in his own funny way.
>(Ed, I hope you get a first impression how stupid any exclusion even of
>the most stupid newbie might be, a fallacy so to speak. Human beings
>have very different strengths. What you techno beasts need is the help
>of people who are _not_ in the business but close enough to know what cc
>program is all about _and_ able to do so-called 'brainstorming' to
>provide you with "new" or notr so new ideas. Know what I mean? Let's
>start that. Life is too short ... I'm waiting with literally minimum 30
>new ideas a tempo. Now think of all the other members too. Do you really
>believe that burnt out older experts will have all the new ideas
>themselves?)
When I look at my database there are over 100 suggestions to improve
my engine and I don't expect that this is different with others
Ed Schroder
who
But, Ed, why not come out with 2 different versions of your program,
one for comp-comp, and one for human-comp?
The comp-comp won't take up that much in resources,
(since you're arguing that less is more in comp vs comp),
and while still having a comp(etitive) comp-comp product out there,
you could also bundle it with another version,
a knowledge-based one that would play wondrous positional chess,
and designed for a serious human's enjoyment/development.
And isn't that where you're heading, anyway? I mean, looking at your own
anti-gm 'function'...
Plus, you could also include a disclaimer with your product, 'educating'
every enduser
out there about the fact that comp-comp is misleading, hurts-your-chess,
blahblahblah,
and is only meant for comp-comp, and nuthin' else...
Eventually the average enduser could become savvy enough to realize that
comp-comp is a
vastly inconclusive pursuit, with no bearing on his/her chess progress.
Why not keep trying? I think your post can be the launchpad to something
very positive,
if we just don't lose hope :)
I think the 'bean-counter' approach is self-destructive in the end.
Knowlege is the way, if the machines are to serve _us_.
(and we'll always have more speed, right ? :-))
Confidently,
"+++++"
PS When is that newer, meaner CsTal going to be released ??
;-)
Ed Schroder wrote in message <36E18390...@rebel.nl>...
who
>>
>>Better divide playing strength into categories and give it a description:
>>
>>. Playing style .....................
>>. Human-alike play .....................
>>. Positional understanding .....................
>>. Positional sacrifices .....................
>>. Tactics .....................
>>. Comp-Comp .....................
>>
>>or whatever is important to know about a chess engine.
>
>Tell me who. Who can judge this ?
>
>I only know Thorsten. And even he talks nonsense at times.
>
>>
>>Maybe (just maybe) when chess journalists start to judge chess
>>programs in another way and emphasize on other points other
>>than comp-comp and tactics chess programmers are willing
>>(forced) to return to the things that really matters and that is
>>adding new and better chess knowledge instead of removing
>>chess knowledge from their programs.
>
>Sure. But you give these people a lot of power. Are they up to it ?
Maybe the real 'expert journalist' must rise from amongst us ...
Not too late (!?)
;-)
>
>Chris Whittington
>
>
>
>
who
Albert Silver wrote in message <36E1A88F...@unisys.com.br>...
>
>
>Chris Whittington wrote:
snip,snip all around
>
>You sound bitter there. I don't think it's that simple. I don't think they
buy
>bean-counters, as you put it, because they can't play chess, and the way
you
>imply it, don't care. True, a weaker player by the very description does
not
>play a great game of chess, and therefore, when confronted with the choice
of
>available programs, cannot simply play them and see which one is stronger
or
>better suited. The player will have to read articles, listen to the experts
or
>at least people he or she believes to be in the know. Yet it is the experts
who
>are declaring so and so program stronger as it outcalculated, or as you put
it,
>'outmined' the other program. Not the weaker players buying them. Blame the
>experts, blame the media. Not the consumers.
Yes, Albert, we need better experts! <seriously, too>
>>
>>
>> I know I can make something that will wipe them out. Really now a
question
>> of whether I can be bothered to do it.
>
>:-) Just dying to have someone say it aren't you? Of course, it may not be
quite
>as interesting coming from a non-chess-programmer (though maybe not,
considering
>your opinion of them), but I'll say it:
>
>I dare you!
Yeah, Chris should go for it! ;-)
just do it, bro !!
>
><snip>
>
> Albert Silver
>
>> >
Chris Whittington
Ed Schroder wrote in message <36E241CF...@rebel.nl>...
>>From: TUESCHEN.MEDIZ...@t-online.de (Rolf Tueschen)
>
>>>I don't think I have the power to do such a thing. I do think that "chess
>>>journalists" is the key word here. If I (actually we) can convince them
>>>they might change their minds and review programs differently.
>>
>>Ed, it's a matter of perspective. From the outside, believe me for one
>>time, it looks different. Of course you set the standards as
>>programmers/business guys. Just one idea. Do you think that "chess
>>journalists" would make difficulties. Didn't you mean "computer chess
>>journalists"? I think Chris said what had to be said. Where do you see
>>for instance one single 2200 cc journalist? Prove me ignorant!
>
>Enrique is. As a result I see very well balanced wordings from his side
>when the subject is playing strength. Like to add Enrique is a comp-comp
>lover still he manages to write good reviews about playing strength so it
>can be done.
>
Now you do yourself exactly what you complain of is done to you. You're
making ad hominem (if complimentary) comments to tack on to a reviewer (who
just happens to be a personal friend of yours) such that 'we' are now
supposed to 'respect' reviews from him more than from elsewhere. Such is the
way myths are created. Tricky, tricky.
You make a review of a reviewer that is probably even more shallow,
nonsensical and meaningless as the same reviewers jibing at me of 18-2 last
year in a pathetic attempt to try and get at me through my program.
If he does that he'll write biased on anything. Does he like the programmer
? Nice worded review. Does he not like the programmer ? Ignore it, or jibe
18-2. This is the autoplayer overnight and go to sleep champion, no ?
If Enrique really understood computer chess programs, or loved them, or
wanted to describe the essential issues involved for 'his' readers, or
whatever, he would have shown it with a senseful review of CSTal. He never
has done - never even tried. If they can't write sensefully on this
particular program, they're no good, they're stuck someplace. It is *the*
test for them. They can't get away with the usual bollocks and platitudes -
they have to step beyond. If they can't, well - the conclusion is obvious.
Chris Whittington
Chris Whittington
Rolf Tueschen wrote in message <7bs6nh$k2t$2...@news05.btx.dtag.de>...
I think you just misunderstood me.
Now maybe I am misunderstanding you. What are you trying to tell me, you
actually want to go and operate CSTal in Paderborn ? Or the created
archetypes are real. Or what ?
Gu-inness is good for you ? I always found it took too long for the froth to
clear and even then it was dark and incomprehensible.
Too many actors spoil the froth. Hey, I can play this game too ! But not
with myself.
Chris Whittington
>
>>Chris Whittington
>
>
jno...@my-dejanews.com
How much would you pay for a , say, 2400-rated computer program?
Not just any 2400, mind you..., but a 2400 that plays like a 2400 in every
way, tactically, positionally, opening, middlegame, endgame, whatnot...
Now, unless you're much stronger than 2400, you'd find such a beast mighty
attractive, wouldn't you say ?
Well, unfortunately, such a program doesn't exist yet.
But, are all possible efforts being made towards achieving such an aim ?
The aim of concentrating on making the engine much stronger exactly where it's
now weak, EVEN IF it comes at the expense of weakening its strong points, i.e.
short-term tactical prowess. Taking a schizoid 3000-rated tactical player, and
improving on its 1800-like positional/strategic understanding, until an
'acceptable', 2600/2200 balance is achieved (as a compromise to the 2400/2400
"ideal"), a _convergence_ of the program's strengths, if you will...
Forgetting for the time being about defeating the WC, or other GMs for that
matter. Just concentrating on producing a TRUE Master-Strength program! With
the more-or-less average player in mind... A program that would play _like_ a
human master in a sensible, recognizable way. (Only a 2400, no more, unless
more is readily feasible, of course)
Now, would this be a worthwhile aim?
Regards,
Jay
In article <HMjE2.38289$rs2.11...@client.news.psi.net>,
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Ed Schroder
>But, Ed, why not come out with 2 different versions of your program,
>one for comp-comp, and one for human-comp?
>
>The comp-comp won't take up that much in resources,
>(since you're arguing that less is more in comp vs comp),
> and while still having a comp(etitive) comp-comp product out there,
>you could also bundle it with another version,
>a knowledge-based one that would play wondrous positional chess,
>and designed for a serious human's enjoyment/development.
>
>And isn't that where you're heading, anyway? I mean, looking at your own
>anti-gm 'function'...
>
>Plus, you could also include a disclaimer with your product, 'educating'
>every enduser
>out there about the fact that comp-comp is misleading, hurts-your-chess,
>blahblahblah,
>and is only meant for comp-comp, and nuthin' else...
>
>Eventually the average enduser could become savvy enough to realize that
>comp-comp is a
>vastly inconclusive pursuit, with no bearing on his/her chess progress.
>
>Why not keep trying? I think your post can be the launchpad to something
>very positive,
>if we just don't lose hope :)
>
>I think the 'bean-counter' approach is self-destructive in the end.
>Knowlege is the way, if the machines are to serve _us_.
>(and we'll always have more speed, right ? :-))
Great post, I loved every word of it.
Did I sound desperate? I hope not because I am not, I only worry about
the negative side effects of giving "comp-comp" too much attention (value)
in the chess press (also internet) it deserves.
As described in previous postings the "too high attention" drives (forces)
chess programmers to add even more tactics and even to remove chess
knowledge from their chess programs because in comp-comp it scores
better.
If this is the case (and it is) the chess press should write about this and
explain their readers what is going on. For the moment I must conclude
they have to wake up first.
As a matter of fact I am really positively surprised about the general
approval on the subject as I expected totally the opposite! and I am
still waiting for good counter arguments.
For Rebel I already have a ready-made solution. Rebel for human-comp,
Chess Tiger for comp-comp. It all fits, I am not willing to take out chess
knowledge, I simply can't as it is against everything I believe in since I
started in 1981.
The big problem I see in this discussion is to convince the chess press
about what we are discussing now as they are obliged to tell their readers,
"it is stronger" or "it is not stronger".
Question: what is stronger :)
Looking at the current chess strength of programs I don't envy the chess
press.
Ed Schroder
Chris Whittington
Phil Innes wrote in message <36E1FBDC...@sover.net>...
>This is the tail-end of a post by Ed Shroeder - to Chris W
>
>positional player????
>
>And the fact that important human-comp tournaments (AEGON,
>Harvard cup) are gone makes comp-comp even more important
>or even worse you can ask yourself the question: what else is left?
>
> always result in the defeat of the computer by humans rated
> 2000. Without the opening book, is any program stronger than
Difficult, at first sight you're wrong, of course. Strip the opening book
out of all the programs and leave them to play humans from the start
position ...... ? They (the prgs) will get themselves in a mess due to lack
of understanding of opening principles, but the game is long, still plenty
of chance for the human to go treading on random mines. Maybe 100 ELO's - I
throw out a wild guess.
But, on second look, try it another way. Give the human the same (or better)
database lookup facility as the program has. No of wins, draws, losses,
which humans have played the line, any recent innovations etc. And not just
ECO, but whole games, just like the big books in programs. So then most
games will result in coming out into the middle game with a well known theme
and strategy, human has, almost by definition a plan of what to do. Even
resulting in just plain 'technical' win. Maybe this is worth 100's of ELOs.
400 ELO even - I throw out more wild guesses.
>
>Thinking all of this over the last months I came more or less to the
>conclusion that we have entered a new area, that adding new extra
>chess knowledge works out counter productive and that progress
>in computer chess will stop or already has been stopped as I can
>not believe that removing chess knowledge is progress.
>
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
>
>So where do we stand now in computer chess?
>
> What do people use their programs for? Storing games, an
> on-line interface, to learn openings, to test tactical skills.
> All this seems constant.
>
> Some time ago Chris down-played my comment about the opening
> book, but Bob seemed to agree with me. I don't understand
> Chris' objection.
It was inherently choleopteristic, pargrammatical psychology with a smithen
of oregano. Comrade.
>
> Ed is very honest in pointing out the basis of current
> development and improvements. But these seem like very small
> increments to me.
>
>And more important, what is the way to go?
>
> Learn why humans play (potentially) better chess. Emulate it.
Right.
>
> Get off your "we the programmers control the thing entirely"
> attitude and revisit the subject by inviting other opinion.
Difficult. The stupid always form Unions to keep dangerous new ideas at bay.
They are also inclined, when the going gets really tough, to form whole new
organisations to keep dangerous ideas out and banned. Also dangerous ideas
are boring. So shut up. And they are off-topic. So shut up. And they take
too many words to read, and since they are so busy. Shut up. Oh, yes, and
they've forgotten what happened it was so long ago. So shut up. And the
email file was lost in the hard-disk crash. So shut up. And anyway nobody
wants to hear this stuff. So shut up. And then there are those who are too
ignorant to understand your language. So shut up.
>
> The crux of the thing is, no matter how fast the linear
> processor is - all positional values are pre-ordained (before
> move 1) by the programmer. These values do not change during
> the game (although value set b, then, set c may be invoked)
> and neither does the program learn anything.
>
> (This has also been a problem in a broader field of robotics.
> Advances were made by interacting with pyschologist, then with
> choleopterists.)
>
Chris Guinness is good for you.
> Phil Innes
>
>
>
>Ed Schroder
>
Ed Schroder
>>Enrique is. As a result I see very well balanced wordings from his side
>>when the subject is playing strength. Like to add Enrique is a comp-comp
>>lover still he manages to write good reviews about playing strength so it
>>can be done.
>>
>
>Now you do yourself exactly what you complain of is done to you. You're
>making ad hominem (if complimentary) comments to tack on to a reviewer (who
>just happens to be a personal friend of yours) such that 'we' are now
>supposed to 'respect' reviews from him more than from elsewhere. Such is the
>way myths are created. Tricky, tricky.
Come on Chris!
It is my personal opinion Enrique is good when the topic is playing
strength. My opinion is just one of the many and has no more value
than the opinion of others.
I am also in many cases in disagreement with Enrique. This doesn't
mean I am always right and vice versa. Computer chess is much too
complicated for that.
Besides Enrique there are of course others as well which I consider
good reviewers when the topic is playing strength but maybe your
point is I shouldn't have mentioned any name at all?
Maybe I should have not, but the question was if there were any writers
above 2200 elo. There are and Enrique is just one of them.
Ed Schroder
Mike Hurd
all the comments expressed here, surely if a chess program could beat a top
grandmaster @40 in 2 Hrs say 10 - 0 or so, wouldn't this same program beat
every other program out there in comp vs comp or am I missing something?
Regards
Mike
Ed Schroder wrote in message <36E2687A...@rebel.nl>...
mclane
>>Paderborn would be the right place for you. Imagine CST (4000) &&
>>CST (300,000) having the 100% score. The ultimate proof :)
>Fine, but for the first time of WCCC they want money $500. And Thorsten is
>banned from the icca, he's my operator, and that cancels me from sending
>him, based on rule '50% of team must be icca member'. Besides, the whole
>thing will be a police state designed to discover malcreant Craftys. I'm
>sick of police states.
i am not sure about my status.
why don't you sell one of your rolls to your neighbour ?
>I've never seen a CSTal review that captures it yet. Either it's sycophantic
>praise from someone who you know knows zilch about it. Or else it's
>nah-nah-nah-nah, 18-2. Or else it gets ignored totally. The journalists are
>the worst of the lot. They know NOTHING.
! :-))
there is a new report in europa-rochade chris.
>I only know Thorsten. And even he talks nonsense at times.
this will cost you at least 2 bear and one currywurst :-)) my master.
best wishes
Chris Whittington
jno...@my-dejanews.com wrote in message <7btpid$od4$1...@nnrp1.dejanews.com>...
Probably. But then they shout 18-2 at you.
Chris Whittington
Chris Whittington
Mike Hurd wrote in message <7btq84$a4q$1...@newnews.global.net.uk>...
>Although I do not write chess programs and perhaps do not fully understand
>all the comments expressed here, surely if a chess program could beat a top
>grandmaster @40 in 2 Hrs say 10 - 0 or so, wouldn't this same program beat
>every other program out there in comp vs comp or am I missing something?
Yes you are. It isn't the same game.
One is chess. The other is blundering around with a feely stick.
One is intelligent, strategic and susceptible to cunning ideas, thoughts,
knowledge. The other is a random game of chance. Whoops, trod on a mine -
hit the take-back key, try again.
Chris Whittington
Giorgio Bertazzo
>The big problem I see in this discussion is to convince the chess press
>about what we are discussing now as they are obliged to tell their readers,
>"it is stronger" or "it is not stronger".
>
>Question: what is stronger :)
Don't underestimate market.
At least 95% of the owners of a modern chess program can be easily
destroyed by their software, and they play mainly bliz games.
If it's free they will download the newest version, but if it's a
commercial program they need a good reason to upgrade to it.
Fritz 5.16 added cbh support. Fritz 5.32 added 32 bit Windows.
Rebel 8,9,10 offered mainly dos,dos and dos.
These were the reasons to switch for many players, not the claims for
a "stronger" program.
If you want to make a mass market winning program add bells and
whistles and a good beginners tutorial and jump to Windows.
This is what Chessmaster did, and can be done better, but this isn't
for one man companies.
If you want to make a winning niche market program forget for a while
comp-comp, jump to Windows and add better analysis and publishing
functions, because a lot of people is seeking for programs able to
analyze and comment their games in the way books and magazine do for
top players games.
I know a lot of players up to 2400 range that use Firtz or Rebel or
Genius to analyze their just played games, mainly to search for
tactical mistakes.
I bought Junior because it delivers better analysis than Fritz, but I
believe that there is room for a lot of improvements both in the
quality of analysis and in its publishing.
I know that these are matters that make the line between playing
programs and databases subtle and confusing...
Chris Whittington
Ed Schroder wrote in message <36E2687A...@rebel.nl>...
>>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>
>>>Enrique is. As a result I see very well balanced wordings from his side
>>>when the subject is playing strength. Like to add Enrique is a comp-comp
>>>lover still he manages to write good reviews about playing strength so it
>>>can be done.
>>>
>>
>>Now you do yourself exactly what you complain of is done to you. You're
>>making ad hominem (if complimentary) comments to tack on to a reviewer
(who
>>just happens to be a personal friend of yours) such that 'we' are now
>>supposed to 'respect' reviews from him more than from elsewhere. Such is
the
>>way myths are created. Tricky, tricky.
>
>Come on Chris!
>
>It is my personal opinion Enrique is good when the topic is playing
>strength. My opinion is just one of the many and has no more value
>than the opinion of others.
>
>I am also in many cases in disagreement with Enrique. This doesn't
>mean I am always right and vice versa. Computer chess is much too
>complicated for that.
>
>Besides Enrique there are of course others as well which I consider
>good reviewers when the topic is playing strength but maybe your
>point is I shouldn't have mentioned any name at all?
>
>Maybe I should have not, but the question was if there were any writers
>above 2200 elo. There are and Enrique is just one of them.
I think I leave it to a Dejas miner to demonstrate that this isn't so.
Nevertheless, you were right in another post. The 'reviewer,
so-called-experts' are a hangover from a few years ago when the programs
were much weaker. Then the 'experts' could understand a bit and write some
sense. But, strength-wise, the programs went right through and beyond them,
so the 'experts' then resorted to autoplaying, sleep and 6.4. Put simply,
program performance and properties are far too complex for them to
comprehend anymore. So now reviews are full of 10.5-9.5 comp-comp, and
platitudes such as 'human-like', or xyz is an 'interesting' program, or some
other way of saying nothing in as long words as possible.
Partly it is also our fault, because we never discuss anything. And, if we
try, we get spammed wtih Crafty this and Crafty that and Crafty the other.
No higher level concepts ever get discussed on the ngs.
Technically, it is all 0x88, my move generator makes 240,000 nps, bitmaps,
what is a node and how do you count them, and 'I wrote some special code to
help my program from playing Bxa7 - trapped bishop'.
Utterly trivial. All they talk about is the mechanics of making a program,
actually the mechanics of making a bean-counter program that does even know
attacks-to, attacks-from or what is and isn't pinned, let alone whether
anything can give check. They are stuck in the frigging coding. Making it
faster. That's where the discussion level is. Thank you Hyatt, mostly. Thank
you programmers of strength mostly 1700.
Every time I see the word bitmap, I release the safety catch on my revolver.
Frigging bitmaps. So someone thought a 64-bit register could map the 64
square chess board. Wow ! Or Doh ! So incredibly trivial.
Or somone thought that a rank, file or diagonal was, at most 8 long, so you
could use 8-bit lookups to tell you stuff. Wow ! I think my friend Bart
Simpson could have come up with that.
Yet this talk of bitmaps is a serious claim to fame, no ? They rant on about
it, like its the biggest deal of all time. When all it is, is a way to
represent the data.
What they then 'do' with this data is never discussed. How it relates to
chess concepts is never discussed. And, the real design question for a chess
program, whether they generate attacks-from, attacks-to and pin information,
or whether they don't, or whether they half do it, or whether they might do
it on the fly - such a primitive design decision, utterly integral to what
type and style of program gets produced - again, never discussed. It's all
just the one way of the Crafty this, Crafty that, and Crafty the other. All
other ways of looking are spammed.
The result, a mass of easy-to-build fast searchers, spamming everybody, and
distorting the entire field.
When Rolf says, he could talk and be useful, if the level was raised. It's
true. If you got beyond bitmaps and basic wheels and cogs of the engine, and
on to what to do with the data that's represented, what type of data should
be represented, what type of data do you need to encapsulate dynamic
knowledge, should a program know this, or that, or the other ? What new
properties can emerge from different ways of searching, or different
knowledge encodement ? All the next set of higher level questions. Then the
people doing the engine design would actually be able to talk to those
people who understood chess.
But, of course, we know this is a pipe-dream. Firstly the programmers are
too bogged down in implementations. And the chess knowledge people aren't
very interested anyway. So forget it.
Chris McGuinness is bad for you.
Rolf Tueschen
>Maybe I should have not, but the question was if there were any writers
>above 2200 elo. There are and Enrique is just one of them.
Ed, with the claim of not knowing his private vita, but I must state
that I don't believe you. Further disclaimer, that I still think that
Enrique is the real name of the guy. :)
Look, it's very improbable that Enrique is a real good chessplayer at
all. I had an in-depth email exchange with him. Now, the first thing I
like to do is asking my correspondant for his most beautiful game he
ever played. Because I enjoy OTB chess so much. Tell you what, Ed, he
answered, that unfortunately he had no games.
And believe me this, here I aggree with Chris, I did NEVER read
something of Enrique that had to do with chessic knowledge. Never. But
again, prove me ignorant and I will apologize.
Also I must add that I'm sure that Enrique is a very smart and educated
guy. There's no doubt about it. But take me by my word. I saw things
written by Enrique that are inconsistent with this smartness and
education. Only this. I wouldn't say that Enrique is a wise man. He was
wrong too many times. And I was the stone of proof for him. Tell you
what. He simply failed. Period.
Enrique was so much wrong, so far away from a decent standpoint of view,
that I must state my serious doubts that Enrique is really an educated
man. Something is wrong in his presentation. Something, other guys here
on the net, perhaps with the same caliber, simply never did do wrong.
Ed, you must also take into consideration those who did not take part in
a overall defamation campaign. And taken Enrique's alleged titles, he
simply failed in the lackmus-test ...
I have a critic against Chris. You surely didn't do ad hominem. If you
stated yur being impressed by Enrique it's well your right to talk
about. But then I must support Chris. Again, I did never read something
substantial from Enrique. All the results from month-long testing with
autoplayer he published in CCC months ago was not really what one would
expect from such a knowledgable man. Even mclane wrote more interesting
stuff about his christmas tournament, IMHO.
So, yes, it completely is not to understand why you think that Enrique
is such an expert of extra class. He's also not able to dicuss. What he
did was mostly throwing his name into the debate and adding a few
bon-mots. The last time he did it was on Faber. But months ago when
Enrique was jolly jumper in CCC he propagated the _same_ nonsense which
he now did oppose.
You see, Ed, the longer we stay here, a certain right of a fresh
approach totally in new clothes is no longer granted. -- Don't take me
wrong. I can talk with Enrique without difficulties. It's a piece of
cake. But don't confuse it with a general acceptance of former
wrong-doings.
Can'tr you understand how Chris must feel in his exclusion as one of
your direct collegues? If you have any time/dedication left you should
think about your possibilities to do Chris something good. Or do you
expect him to beg you? Ed, try to understand me. Certain things must be
corrected. And the best thing to do is NOT to write a sudden off-topic
declaration but to do it on-topic. The name of Enrique, in the way you
presented it was another slap in the neck for Chris. Not a
positive/productive approach.
Finally it's understood that I'm not involved in all that personally. :)
>Ed Schroder
Jesper Antonsson
>Mike Hurd wrote in message <7btq84$a4q$1...@newnews.global.net.uk>...
>>Although I do not write chess programs and perhaps do not fully understand
>>all the comments expressed here, surely if a chess program could beat a top
>>grandmaster @40 in 2 Hrs say 10 - 0 or so, wouldn't this same program beat
>>every other program out there in comp vs comp or am I missing something?
>
>Yes you are. It isn't the same game.
>
>One is chess. The other is blundering around with a feely stick.
>
>One is intelligent, strategic and susceptible to cunning ideas, thoughts,
>knowledge. The other is a random game of chance. Whoops, trod on a mine -
>hit the take-back key, try again.
Your posts are abstract and are very coloured by emotions. Not a good
combination; it's hard to extract any real information. Well, I suspect
even your program is a tactical monster by human standards, but quite bad
positionally, even if it is somewhat better than some other engines. So,
you too are blundering around with a "feely stick". If there is any
difference, it's about very small nuances.
I will try to make Ed happy, defending an opposite position on fast/smart
in comp-comp. :-)
Since programs like Hiarcs, with low NPS counts, are successful in
comp-comp, I don't believe you have to rip out knowledge and make it fast
and dumb to win comp-comp. If you do have to do this, your eval terms (or
extensions) are most likely objectively unsound in the first place.
Whenever you add an eval term, it should yield better results on average,
even against computers, than before, or else the eval term is too expensive
to compute or perhaps even unsound in itself (or in relation to the other
existing eval terms).
If your eval terms let you play a really good positional game, then it
won't fall for many pitfalls when playing other computers, rather it
would more often get good enough positions so that a faster searcher just
suddenly finds itself in a lost position and can't do anything about it.
Weren't this what Ed claimed happened in the one handicapped game against
Crafty?
Of course, it is possible to tune a bit to be more successful against
computers, but this "I have to rip everything out" I don't believe until
I get convincing evidence.
And when Chris self-righteously talk about bean-counters, that others are
dumb, and that reviews of CSTal is the ultimate test for determining
the worth of the reviewing chess journalists, I frankly get a bit sick...
Jesper
Robert Hyatt
: Ed Schroder wrote in message <36E2687A...@rebel.nl>...
I think that is _total_ baloney. IE I believe I participated in a long
thread with you about parallel search. That is a pretty "high-level"
concept that seemed to interest you.
We have also had discussions here (not necessarily you, but lots of
others) about move ordering (Hsu joined in the SEE vs MVV/LVA discussion
a few years ago). We have discussed various evaluation ideas. Search
extensions ad nauseum (singular extensions, threat extensions, and who
knows what else.
And I don't believe that this newsgroup gets "spammed" by Crafty does this
and Crafty does that. Quite the opposite. I'm one of the _few_ programmers
here that _will_ tell what I do and how I do it. So rather than getting
'spammed' by me, the NG gets "stonewalled" by folks such as yourself. There
_is_ a difference.
: Technically, it is all 0x88, my move generator makes 240,000 nps, bitmaps,
: what is a node and how do you count them, and 'I wrote some special code to
: help my program from playing Bxa7 - trapped bishop'.
: Utterly trivial. All they talk about is the mechanics of making a program,
: actually the mechanics of making a bean-counter program that does even know
: attacks-to, attacks-from or what is and isn't pinned, let alone whether
: anything can give check. They are stuck in the frigging coding. Making it
: faster. That's where the discussion level is. Thank you Hyatt, mostly. Thank
: you programmers of strength mostly 1700.
: Every time I see the word bitmap, I release the safety catch on my revolver.
: Frigging bitmaps. So someone thought a 64-bit register could map the 64
: square chess board. Wow ! Or Doh ! So incredibly trivial.
: Or somone thought that a rank, file or diagonal was, at most 8 long, so you
: could use 8-bit lookups to tell you stuff. Wow ! I think my friend Bart
: Simpson could have come up with that.
: Yet this talk of bitmaps is a serious claim to fame, no ? They rant on about
: it, like its the biggest deal of all time. When all it is, is a way to
: represent the data.
I haven't seen it talked about "like it is the biggest deal of all time." It
is only 25+ years old, dating back to kaissa and chess 4.0, so it isn't exactly
new. Rotating them is new. But it is an evolutionary enhancement.
: What they then 'do' with this data is never discussed. How it relates to
: chess concepts is never discussed. And, the real design question for a chess
: program, whether they generate attacks-from, attacks-to and pin information,
: or whether they don't, or whether they half do it, or whether they might do
: it on the fly - such a primitive design decision, utterly integral to what
: type and style of program gets produced - again, never discussed. It's all
: just the one way of the Crafty this, Crafty that, and Crafty the other. All
: other ways of looking are spammed.
Total baloney. Bruce has discussed _his_ approach. I have discussed mine.
Spam != information. You are defining spam in a silly way... where you
provide +zero+ information about how _you_ do something, and then label as
'spam' that which _anyone_ else provides in that silence.
Not only do I explain _exactly_ what I am doing, I have provided source code
to _show_ what I am doing, whether it is 'good or bad or experimental'. You
seem to have a real problem with 'crafty' since you refer to it negatively
many times. Which is a change from the way _most_ of us discuss other
programs. IE instead of talking about what you do that you think is done
well, you'd rather talk about what _I_ do and why you think it is bad.
Seems like a strange perspective... IMHO.
: The result, a mass of easy-to-build fast searchers, spamming everybody, and
: distorting the entire field.
There is that 'spamming everybody' nonsense again. You are reading a
different newsgroup than I am, it seems.
--
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
Robert Hyatt
: jno...@my-dejanews.com wrote in message <7btpid$od4$1...@nnrp1.dejanews.com>...
: Chris Whittington
Then improve the program so that the 18-2 result doesn't happen, rather than
complaining when it does and trying to justify why it happened. The game
of chess is about _winning_ and _only about_ winning. Last time I looked,
no one could win a tournament by playing beautiful chess. They have to win
games...
Robert Hyatt
: This is the tail-end of a post by Ed Shroeder - to Chris W
: To compete in this area you have to make your program a weaker
: positional player????
: And the fact that important human-comp tournaments (AEGON,
: Harvard cup) are gone makes comp-comp even more important
: or even worse you can ask yourself the question: what else is left?
: Human comp tournaments played by normal chess rules would
: always result in the defeat of the computer by humans rated
: 2000. Without the opening book, is any program stronger than
: 1900, even running at 500Mhz?
Yes they are. _significantly_ stronger. The primary advantage of the
book is to avoid repetitious behavior. But even that has been eliminated
in at least a few programs. I don't think the book is a huge advantage at
all, as I have seen lots of programs that would follow opening theory without
any operational book at all. Many current programs _might_ not do this,
since it is easy for the author to rely on the book and ignore any sort of
'opening evaluation'. But most worry about openings like 1. h3 and 2. a3
and put some sort of developmental evaluation into their program to avoid
getting killed when the opponent does that.
: Thinking all of this over the last months I came more or less to the
: conclusion that we have entered a new area, that adding new extra
: chess knowledge works out counter productive and that progress
: in computer chess will stop or already has been stopped as I can
: not believe that removing chess knowledge is progress.
: The fact that Program_X scores better against Program_Y doesn't
: (necessarily) mean that X is better than Y. X just scores better
: against Y and that's all. Maybe Y scores a lot better against humans
: than X and Aegon certainly has proven that.
: So where do we stand now in computer chess?
: What do people use their programs for? Storing games, an
: on-line interface, to learn openings, to test tactical skills.
: All this seems constant.
: Some time ago Chris down-played my comment about the opening
: book, but Bob seemed to agree with me. I don't understand
: Chris' objection.
: Ed is very honest in pointing out the basis of current
: development and improvements. But these seem like very small
: increments to me.
: And more important, what is the way to go?
: Learn why humans play (potentially) better chess. Emulate it.
: Get off your "we the programmers control the thing entirely"
: attitude and revisit the subject by inviting other opinion.
: The crux of the thing is, no matter how fast the linear
: processor is - all positional values are pre-ordained (before
: move 1) by the programmer. These values do not change during
: the game (although value set b, then, set c may be invoked)
: and neither does the program learn anything.
Not totally true... there are lots of 'second-order' evaluation things
done in programs... ie things that 'interrelate' like king exposure and
attracting pieces toward the king.
I agree that programs don't 'adjust' these weights based on game results
so that they 'learn' of course... But the set of 'weights' is so non-
linear that it would be very difficult for a human to 'use' this information
in a match, for example. IE if you could detect the precise 'bishop pair'
bonus, you might use this to exploit the program (Roman did this against me
for a long time, continually complaining about 'it is too high', 'now it is too
low'. It is a dynamic term (now) that includes two bishops, plus bishop
mobility, pawn structure, and so forth. Which means that in position A it
might be value X, but in position B it will be different. So exploitation is
not so easy...
: (This has also been a problem in a broader field of robotics.
: Advances were made by interacting with pyschologist, then with
: choleopterists.)
: Phil Innes
: Ed Schroder
Chris Whittington
Jesper Antonsson wrote in message
<4D6FBECA0F2784E3.E26533A0...@library-proxy.airnews.ne
t>...
>"Chris Whittington" ch...@cpsoft.demon.co.uk wrote:
>
>>Mike Hurd wrote in message <7btq84$a4q$1...@newnews.global.net.uk>...
>>>Although I do not write chess programs and perhaps do not fully
understand
>>>all the comments expressed here, surely if a chess program could beat a
top
>>>grandmaster @40 in 2 Hrs say 10 - 0 or so, wouldn't this same program
beat
>>>every other program out there in comp vs comp or am I missing something?
>>
>>Yes you are. It isn't the same game.
>>
>>One is chess. The other is blundering around with a feely stick.
>>
>>One is intelligent, strategic and susceptible to cunning ideas, thoughts,
>>knowledge. The other is a random game of chance. Whoops, trod on a mine -
>>hit the take-back key, try again.
>
Of course. I'm alive.
>Not a good
>combination; it's hard to extract any real information.
Sorry, I'll try to be a database.
>Well, I suspect
>even your program is a tactical monster by human standards, but quite bad
>positionally, even if it is somewhat better than some other engines. So,
>you too are blundering around with a "feely stick". If there is any
>difference, it's about very small nuances.
We can discuss the feely sticks. Interesting idea. Mine sees round corners,
and probes in more than one dimension.
Feely sticks and minefields. You like the idea ?
>
>I will try to make Ed happy, defending an opposite position on fast/smart
>in comp-comp. :-)
>
>Since programs like Hiarcs, with low NPS counts, are successful in
>comp-comp, I don't believe you have to rip out knowledge and make it fast
>and dumb to win comp-comp. If you do have to do this, your eval terms (or
>extensions) are most likely objectively unsound in the first place.
>Whenever you add an eval term, it should yield better results on average,
>even against computers, than before, or else the eval term is too expensive
>to compute or perhaps even unsound in itself (or in relation to the other
>existing eval terms).
Suppose adding an eval term costs more time - so you throw it out right,
pragmatic grounds ?
And your next eval term idea, and the next. All thrown out for good
pragamatic reasons. It socred lower after 10,000 games or whatever is your
metric.
But suppose the weight of ALL those eval terms, combined together yielded
somehting really powerful. You'ld never know ?
If they took off exponential ? Only by BELIEVING can you find that.
Pragmatism means no progress off the plateau.
>
>If your eval terms let you play a really good positional game, then it
>won't fall for many pitfalls when playing other computers, rather it
>would more often get good enough positions so that a faster searcher just
>suddenly finds itself in a lost position and can't do anything about it.
>Weren't this what Ed claimed happened in the one handicapped game against
>Crafty?
I have lots of experience of this. Knowledge will always get better
position. nps is fairly irrelevant for that. nps will, can, may still
wriggle out. I think the critical comes at about 4 plies extra search.
>
>Of course, it is possible to tune a bit to be more successful against
>computers, but this "I have to rip everything out" I don't believe until
>I get convincing evidence.
>
>And when Chris self-righteously talk about bean-counters, that others are
>dumb, and that reviews of CSTal is the ultimate test for determining
>the worth of the reviewing chess journalists, I frankly get a bit sick...
Good. Take it all the way and vomit.
CSTal *is* the test for them. Its complex and its different. If they don't
try to understand what's going on, they should quit reviewing.
Chris Whittington
>
>Jesper
Chris Whittington
Robert Hyatt wrote in message <7bu67t$7ne$1...@juniper.cis.uab.edu>...
Yes. It interested me. But it's low level, about programming technicalities.
Chess or not chess, made no difference. We could have been talking about
parallel chocolate bar production.
>
>We have also had discussions here (not necessarily you, but lots of
>others) about move ordering (Hsu joined in the SEE vs MVV/LVA discussion
>a few years ago).
Again, a mere technicality. Just about programming and move ordering.
>We have discussed various evaluation ideas.
Seriously rare.
>Search
>extensions ad nauseum (singular extensions, threat extensions, and who
>knows what else.
Again non-demain specific technicalities. Not chess. Things that interest
programmers, not things that interest chess knowledge people, other than to
get them working and have done with it. Then on to higher things.
>
>And I don't believe that this newsgroup gets "spammed" by Crafty does this
>and Crafty does that. Quite the opposite. I'm one of the _few_
programmers
>here that _will_ tell what I do and how I do it. So rather than getting
>'spammed' by me, the NG gets "stonewalled" by folks such as yourself.
There
>_is_ a difference.
Except you tell us over and over and over, and everyone agrees with you. It
makes for huge mass-agreement-universal monotony. There's only one of me.
There _is_a difference.
Its a technical frippery. We know it. Maybe it should be in a FAQ somewhere.
But not brought up again and again.
>
>: What they then 'do' with this data is never discussed. How it relates to
>: chess concepts is never discussed. And, the real design question for a
chess
>: program, whether they generate attacks-from, attacks-to and pin
information,
>: or whether they don't, or whether they half do it, or whether they might
do
>: it on the fly - such a primitive design decision, utterly integral to
what
>: type and style of program gets produced - again, never discussed. It's
all
>: just the one way of the Crafty this, Crafty that, and Crafty the other.
All
>: other ways of looking are spammed.
>
>Total baloney. Bruce has discussed _his_ approach. I have discussed mine.
>Spam != information. You are defining spam in a silly way...
I'm defining spam as huge quantities of repetitive grunge that I've had
enough of, but can't avoid.
>where you
>provide +zero+ information about how _you_ do something, and then label as
>'spam' that which _anyone_ else provides in that silence.
There isn't anybody here to talk to about it. Shame but true.
>
>Not only do I explain _exactly_ what I am doing, I have provided source
code
>to _show_ what I am doing, whether it is 'good or bad or experimental'.
Right. But you are the newsgroup. Do you get it ? Not that you do it, that's
good. But that it's the dominating paradigm here. Impossible to escape, and
impossible to move on from, like sticky mud.
>You
>seem to have a real problem with 'crafty' since you refer to it negatively
>many times. Which is a change from the way _most_ of us discuss other
>programs. IE instead of talking about what you do that you think is done
>well, you'd rather talk about what _I_ do and why you think it is bad.
>Seems like a strange perspective... IMHO.
>
As we both agreed some time ago, you need somene to keep you under control.
It's my turn every now and again.
>
>: The result, a mass of easy-to-build fast searchers, spamming everybody,
and
>: distorting the entire field.
>
>There is that 'spamming everybody' nonsense again. You are reading a
>different newsgroup than I am, it seems.
>
Well, writing to a different one. But then you had something to do with
that, no ?
Chris Whittington
Chris Whittington
Robert Hyatt wrote in message <7bu6bg$7ne$2...@juniper.cis.uab.edu>...
>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
>
>: jno...@my-dejanews.com wrote in message
Doesn't when we play single games.
> rather than
>complaining when it does and trying to justify why it happened. The game
>of chess is about _winning_ and _only about_ winning.
Bollocks. Bollocls. Bollocks.
> Last time I looked,
>no one could win a tournament by playing beautiful chess.
Could you tell ?
Chris Whittington
>They have to win
>games...
>
>
Albert Silver
> >>well in comp-comp and in these test-sets.
>
>
> No idiots, but a lack of sufficient elo maybe?
>
> >. Human-alike play .....................
> >. Positional understanding .....................
> >. Positional sacrifices .....................
> >. Tactics .....................
> >. Comp-Comp .....................
It's very complicated. Of course you are right in that without sufficient
chess strength no proper evaluation on a personal basis can be done. Still,
even with a Grandmaster judging your program it will be complicated. Give a
grandmaster copies of Rebel 8 and Rebel 9. Ask him to make a detailed
comparison in playing strength and style. I have no doubt he will be very
hard pressed to give you something conclusive. How does he find out if 9 is
40-50 points stronger than 8? How do you do that with human players? How do
you explain what it is that player A is doing that gives him a rating of 2440
as opposed to player B with 2480? Does one write something like has a
slightly better feel (it would be funny reading that a program has a 'feel')
for positions, etc...? With no concrete examples to fall upon such as tests,
positions, and so forth, all reviews will end up saying exactly that as it
has become a guessing game. The company claims the program is 40-50 points
stronger so let's see if we can discover what it is that brought about this
increase, if there is indeed an increase. I suspect that the ONLY way to get
a proper review in this case is with the help of the programmer. Give the
reviewers a press kit so to speak. Tell them what it is you have introduced
and modified. Although this may produce some biased results, it will also
help the reviewer know what to look for. I reviewed a French film called 'The
Hairdresser's Wife' some years back. Naturally, I hardly 'needed' the press
kit that came along in order to review the film. Yet without it my review
would not have been as thorough for the kit revealed details on the origin of
the project, the filming, etc...
Albert Silver
>
> >they have tried in css, you know what happened to hiarcs ??
>
> I think the approach (is) was good, the contents were horrible I agree.
>
> Ed Schroder
>
>
-----------== Posted via Deja News, The Discussion Network ==----------
Chris Whittington
Robert Hyatt wrote in message <7brk4b$dj1$1...@juniper.cis.uab.edu>...
>Ed Schroder <in...@rebel.nl> wrote:
>:>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>
>:>Search (simple material chess evaluation over a tree) will lead to
>: emergent
>:>properties. Some chess knowledge will emerge quickly (like don't leave
>: your
>:>king and queen in a knight fork position). Some will emerge slowly
>: (maybe
>:>like your queen development). And some effectively never. Where
>: quickly,
>:>slowly and never refer to the size of the search space.
>
>: [ snip ]
>
>:>The fact that this process of fire control doesn't work is masked by
>: the
>:>fact that programs keep getting faster. More generalised encapsulations
>: of
>:>higher-level knowledge are needed - that's where I spend (some of) my
>: time.
>:>
>:>Chris Whittington
>
>: Chris,
>
>: What do you think about the statement, "Chess in 98% tactics", this
>: limited to computer-computer games area, let me explain a bit...
>
>: I remember the (old) Hitech experiment of Berliner. He took the Hitech
>: program removed important chess knowledge but left the very basic
>: chess knowledge and called this version Hitech_Low (HL) and his
>: original version HH.
>
>: Then matches were played with HL and HH at several ply-depths.
>
>: It showed up that HL (8-ply) lost from HH (8-ply) but that HL (9-ply)
>: vs HH (8-ply) was about equal and HH (8-ply) was crushed when
>: it played HL (10-ply).
>
>: The conclusion more or less was: (extra) chess knowledge is just
>: worth one ply.
>
>: Especially the last years I have seen the following happen:
>: . make the program as fast as possible (NPS)
>: . this in combination with a low branch factor (null-move)
>: . add only the very basic chess knowledge (piece-square tables etc)
>: . make the program aggressive to ensure tactic complications in games.
>
>
>: As a result (also because of today's fast Pc's) you get your program
>: on incredible deep ply-depths. 13 and 14 plies in the middle game are
>: no exceptions anymore.
>
>: What if a program with a lot of chess knowledge (having a low
>: NPS because of that) is faced with such a "fast searcher" AND the
>: HH_HL principle is true. It will lose no? The "fast searcher" naturally
>: comes 3-4 plies deeper because of a smart search and the HH_HL
>: principle ensures victory after victory against a low NPS program.
>
>: True sofar?
>
>: What do you think of the following statement of mine:
>
>: In comp-comp a smart search is worth 200-300 ELO points.
>
>: IMO many programmers have done the following the last years:
>: - remove (!!) chess knowledge from the program to get a high NPS;
>: - improved selected search (smart pruning) to get a low branch factor;
>: - make the program more aggressive to ensure tactics in comp-comp games;
>
>: As a result the programs do a lot better now in the comp-comp area.
>
>: Is this the end of adding new chess knowledge to our programs?
>
>: It bothers me. To compete in the comp-comp area you have to
>: remove chess knowledge???
>
>: The bad news is, it apparently seems to work!
>
>: This subject has been my attention since a couple of months now.
>: It's not that simple as you have for instance Hiarcs and Chessmaster
>: which are "slow searchers" (a very low NPS) and they both are doing
>: fine in comp-comp.
>
>: However if you take a closer look both programs seems to be
>: specialized in tactics unless the fact they have a low NPS.
>
>: The last months I have worked on Rebel adding tactics, more pruning,
>: a lower branch factor. As a result it performs better in comp-comp but
>: I definitely see a drop in the positional play. The quality of the
>: returned
>: moves is simply less, still it is a better comp-comp fighter now.
>
>: What to do?
>
>: magazines (or internet) because it is compared with other chess
>: programs and the score it gained against them.
>
>: positional player????
>
>: And the fact that important human-comp tournaments (AEGON,
>: Harvard cup) are gone makes comp-comp even more important
>: or even worse you can ask yourself the question: what else is left?
>
>: conclusion that we have entered a new area, that adding new extra
>: chess knowledge works out counter productive and that progress
>: in computer chess will stop or already has been stopped as I can
>: not believe that removing chess knowledge is progress.
>
>: The fact that Program_X scores better against Program_Y doesn't
>: (necessarily) mean that X is better than Y. X just scores better
>: against Y and that's all. Maybe Y scores a lot better against humans
>: than X and Aegon certainly has proven that.
>
>: So where do we stand now in computer chess?
>
>
>
>There is an alternative. Do as I try to do... Ignore computer vs
>computer games unless you see something you are consistently doing wrong
>in those games. I'm not hung up on trying to beat every program by
>being as fast as possible. I have gotten _slower_ the past 3 years, not
>faster, and I think that direction is perfectly ok. I've said before
>that _far_ too much attention and emphasis has been placed on SSDF
>results, which has led to today's situation.
I think you're dead right. And you've been saying it all along to the
Europeans. You were right about SSDF testing, and right about the way it
perverted progress.
But.
Server games have their own perversion. Most are blitz or fast time
controls, bullet even. This counts against knowledge and in favour of fast
searchers.
Also you get literally thousands of games. Thousands of games means a
tendency not to pay attention to the game content, instead the mass results.
Or the up/down elo movement.
So again, because of the easy wins against even strong humans at blitz (too
many mines), the tendency is to go for speed. Or low level knowledge.
>I think it much more interesting
>to continue what I've been doing for along time: trying to find ways to
make
>these 'anti-computer' strategies backfire. Because once you learn how to
>'break anti-computer' you also learn how to _play_ it. And there would be
>nothing to prevent you from using that against other computers, of course.
:)
I still maintain you didn't bite the bullet and encode any dynamic
knowledge. It's interesting to watch you, Bob. Because once you were the
bean-counter par excellence. Then you listened to all the attacks, and you
started to move, but I think it was never 'all the way'. So now you inhabit
a kind of limbo land, neither knowledged, nor proper bean counter. We need
to shake you about a bit more :)
Before you gyrate through the ceiling, thios is how I know. If, you were
going for knowledge, you'ld geberate all attack data. But instead, you still
do it on the fly. This is a disincentive to you to use the data, because
making it costs you so much. Neither one thing nor the other.
Chris Whittington
>
>The hardest problem I am looking at is the book. It is hard to realize
that
>when you play move 2 and 3 in some opening, that you are committing to
either
>locking the pawns up 10 moves further along, or else you are going to
really
>have to compromise your position to avoid that. And I don't see anyone
able
>to handle this yet, other than by hand-tuning. I have 'partially' solved
the
>inability to understand locked up pawns... but only partially, and I still
>have the problem of committing to a locked pawn structure without knowing
it,
>because of the book line chosen.
>
>BTW, note that "LowTech" in Berliner's paper was not _stupid_. It still
>did a _full_ pawn structure evaluation, but after that only had a weak
>centralize-the-pieces eval. However, it _might_ be that the pawn structure
>analysis was the most important thing he did. A poorly placed knight can
>be moved, while a weak pawn remains weak forever.
>
>His results might be a little harder to quantify when you factor that in.
Chris Whittington
Albert Silver wrote in message <7bud9u$7ho$1...@nnrp1.dejanews.com>...
>In article <36E24148...@rebel.nl>,
Right, right, right, right and right.
>How does he find out if 9 is
>40-50 points stronger than 8?
Doesn't matter anymore. This has become unimportant.
>How do you do that with human players?
You don't. You don't watch Linares to see if one guy goes up by 30 and
another down. Do you ? You watch it because Kasparov plays a whole new
exciting way. Who cares about the grade ? Only those who need a number to
qualify with.
>How do
>you explain what it is that player A is doing that gives him a rating of
2440
>as opposed to player B with 2480?
Doesn't matter.
>Does one write something like has a
>slightly better feel (it would be funny reading that a program has a
'feel')
>for positions, etc...? With no concrete examples to fall upon such as
tests,
>positions, and so forth, all reviews will end up saying exactly that as it
>has become a guessing game.
Right. So reviewers need a whole new kit of understanding. Now it is hard
work for them. Should be a challenge.
>The company claims the program is 40-50 points
>stronger so let's see if we can discover what it is that brought about this
>increase, if there is indeed an increase. I suspect that the ONLY way to
get
>a proper review in this case is with the help of the programmer.
Probably. But we lie. And a lot of programmer thought never gets out, it
gets turned into woffle by resellers and publishers.
Bob does it right by talking about his code. But we haven't got away, over
and above programming issues. There's nothing about chess. We don't have the
language. We don't have much desire because its so competitive. And we all
hate each other really :)
>Give the
>reviewers a press kit so to speak. Tell them what it is you have introduced
>and modified.
Then they just make lists. And still don't understand. And the readers don't
understand the lists.
>Although this may produce some biased results, it will also
>help the reviewer know what to look for.
Thorsten has told time and time again what to look for in CSTal. Doesn't
seem to make much difference. Actually the only person I noticed making
informed and revealing comment is Mr Bruce, ages ago. But nobody noticed.
Maybe it's all too complex and difficult now. And they seek answers to the
wrong questions.
Chris Whittington
Robert Hyatt
: Robert Hyatt wrote in message <7bu6bg$7ne$2...@juniper.cis.uab.edu>...
:>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
:>
:>: jno...@my-dejanews.com wrote in message
:>
:>
:>Then improve the program so that the 18-2 result doesn't happen,
: Doesn't when we play single games.
:> rather than
:>complaining when it does and trying to justify why it happened. The game
:>of chess is about _winning_ and _only about_ winning.
: Bollocks. Bollocls. Bollocks.
:> Last time I looked,
:>no one could win a tournament by playing beautiful chess.
: Could you tell ?
: Chris Whittington
Absolutely yes. And I'd rather win a game beautifully myself. But
when it gets down to 'it', I'd rather win, period. IE the choice
between playing a wild sac and going down in flames or slowly winning
a KPP vs KP ending is not much of a choice, IMHO.
Robert Hyatt
: I think you're dead right. And you've been saying it all along to the
: Europeans. You were right about SSDF testing, and right about the way it
: perverted progress.
: But.
: Server games have their own perversion. Most are blitz or fast time
: controls, bullet even. This counts against knowledge and in favour of fast
: searchers.
I disagree. "fast" programs get eaten alive at blitz, if they don't have
a pretty strong notion of 'king safety'. Because they can't search their
way out of a g4/h4/g5/h5 sort of attack at fast time controls, yet humans
know how to play those games blindfolded. A fast program will shred the
lower-rated players, but on a server, it will get shredded by 2000-2300
players that know how to attack. My first year on the chess servers saw
me completely re-think king safety several times. And once you get past
the 2000-2300 players, you rise into the IM players, and _they_ know much
more about attacking. And then there are the GM guys waiting once you solve
most of the IM 'problems'.
IE I'd much rather take a fast program and play 40/2hr than I would blitz.
Because if you are fast, you depend on search, and at blitz, search can find
shallow tactics, but not the deep stuff that you have to handle...
: Also you get literally thousands of games. Thousands of games means a
: tendency not to pay attention to the game content, instead the mass results.
: Or the up/down elo movement.
This can be a problem. I tend to focus only on losses. And I simply play
thru the games, letting the positional 'errors' sink in mentally, and then
review for 'common themes'. IE the Bxa7 problem. Or outside passed pawns.
or preserving the bishop pair but accepting weak pawns in return. Once I
find a 'common theme' the fix usually suggests itself.
: So again, because of the easy wins against even strong humans at blitz (too
: many mines), the tendency is to go for speed. Or low level knowledge.
"Low level" is not a real precise term. But in blitz, some kind of knowledge
is absolutely essential. IE there has been example after example of a "good
program" that gets shredded on the servers because even though it plays good
positional chess, it doesn't understand king safety and gets killed. I won't
argue 'low-level vs high-level' since I'm not sure there is a precise definition
that would lend itself to argument...
But I don't believe a piece/square program (simple eval) can survive on a
server today. We have _lots_ of new programs running, and every one runs into
the same problems... and they follow the same path of 'getting smarter about
protecting the king'.
:>I think it much more interesting
:>to continue what I've been doing for along time: trying to find ways to
: make
:>these 'anti-computer' strategies backfire. Because once you learn how to
:>'break anti-computer' you also learn how to _play_ it. And there would be
:>nothing to prevent you from using that against other computers, of course.
: :)
: I still maintain you didn't bite the bullet and encode any dynamic
: knowledge. It's interesting to watch you, Bob. Because once you were the
: bean-counter par excellence. Then you listened to all the attacks, and you
: started to move, but I think it was never 'all the way'. So now you inhabit
: a kind of limbo land, neither knowledged, nor proper bean counter. We need
: to shake you about a bit more :)
I'm not quite sure what you mean. Cray Blitz was the penultimate 'bean
counter'. We considered a score of +.35 to be nearly winning because the
positional scores were very small in most cases. In Crafty, when I started,
I started with the idea that I wanted to see if that was essential or if it
could be eliminated. I have, from day 1, seen scores of +1.5 with material
equal, (or -1.5 for that matter). I'm not sure it is 'smart', but I found
a long time ago that if I don't give something a chance and try to make it
work, then lo' and behold, it _won't_ work. As a result, I sacrifice pawns
for positional compensation. I sacrifice the exchange here and there. I
sacrifice a piece to open up the opponent's king. Whether this is good or
bad is debatable.
But from the beginning, crafty has been an attempt to 'learn'. I felt I
had 'done my time in the trenches' competing with Cray Blitz. Crafty is
_nothing_ more than a brand new effort, starting from scratch, so that I
can answer a lot of questions I wanted to ask 10-20 years ago. I don't
claim what I do is right or wrong. I do claim that it 'answers questions'.
We could argue all day about checks in the q-search. At present, I want to
know if not doing them is feasible. So far, I believe so. But I am looking,
and I experiment with this from time to time. Ditto things such as singular
extensions, threat extensions, evaluation terms like 'outside passed pawns' (I
can't count the number of programs that used to lose endgames to Crafty before
they decided to add this piece of knowledge.)
I always want to go fast. Because faster = better when I talk about one
program. But I don't go faster by taking things out unless what I remove is
not working well. I expect to continue to get slower and slower. The current
king safety / attack stuff is an example. It is playing better, but it is
playing slower, and it will continue to get slower...
: Before you gyrate through the ceiling, thios is how I know. If, you were
: going for knowledge, you'ld geberate all attack data. But instead, you still
: do it on the fly. This is a disincentive to you to use the data, because
: making it costs you so much. Neither one thing nor the other.
That's the point of what I do. I can do anything. And when that 'anything'
turns out to be 'good' then I have to figure out how to do it 'fast'. I
started out with full attack info. And found I wasn't using it. Maybe I
will re-visit that at some point...
: Chris Whittington
Richard Heldmann
explain why IBM dismantled Deep Blue and sold off the parts.
Rick Heldmann
Phil Innes
> Human comp tournaments played by normal chess rules would
> always result in the defeat of the computer by humans rated
> 2000. Without the opening book, is any program stronger than
> 1900, even running at 500Mhz?
Difficult, at first sight you're wrong, of course. Strip the opening
book
out of all the programs and leave them to play humans from the start
position ...... ? They (the prgs) will get themselves in a mess due to
lack
of understanding of opening principles, but the game is long, still
plenty
of chance for the human to go treading on random mines. Maybe 100 ELO's
- I
throw out a wild guess.
That's a strong (human) players' response. But really, you
seem to agree with this - and becuase I said "if we play by
the rules" then human/comp competition will be forever limited
to sub 2000 play.
But, on second look, try it another way. Give the human the same (or
better)
database lookup facility as the program has.
Hah! Okay - like advanced chess a la Kasparov.
This is not pushing the envelop, we are repackaging to suit
computers' limitiations.
No of wins, draws, losses,
which humans have played the line, any recent innovations etc. And not
just
ECO, but whole games, just like the big books in programs. So then most
games will result in coming out into the middle game with a well known
theme
and strategy, human has, almost by definition a plan of what to do. Even
resulting in just plain 'technical' win. Maybe this is worth 100's of
ELOs.
400 ELO even - I throw out more wild guesses.
>
>Thinking all of this over the last months I came more or less to the
>conclusion that we have entered a new area, that adding new extra
>chess knowledge works out counter productive and that progress
>in computer chess will stop or already has been stopped as I can
>not believe that removing chess knowledge is progress.
>
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
>
>So where do we stand now in computer chess?
>
> What do people use their programs for? Storing games, an
> on-line interface, to learn openings, to test tactical skills.
> All this seems constant.
>
> Some time ago Chris down-played my comment about the opening
> book, but Bob seemed to agree with me. I don't understand
> Chris' objection.
It was inherently choleopteristic, pargrammatical psychology with a
smithen
of oregano. Comrade.
Douche, mon brave. Quel vowels! (see the vowels on 'un!)
What you are meaning to say - but are too shy - is that Bob is
right, right?
>
> Ed is very honest in pointing out the basis of current
> development and improvements. But these seem like very small
> increments to me.
>
>And more important, what is the way to go?
>
> Learn why humans play (potentially) better chess. Emulate it.
Right.
>
> Get off your "we the programmers control the thing entirely"
> attitude and revisit the subject by inviting other opinion.
Difficult. The stupid always form Unions to keep dangerous new ideas at
bay.
Steady-on, big fella! This is the grand reunion pre-millenial
breakthrouh bash.
I think two things are going on:
Your prog, ChickensThongs has some mysterious benefit, what
you have put into it by way of your players' mind.
Meanwhile, Bob points out that there are limitations to this
kind of development, i.e. to what degree is the CSTal
advantage extendible?
They are also inclined, when the going gets really tough, to form whole
new
organisations to keep dangerous ideas out and banned. Also dangerous
ideas
are boring. So shut up. And they are off-topic. So shut up. And they
take
too many words to read, and since they are so busy. Shut up. Oh, yes,
and
they've forgotten what happened it was so long ago. So shut up. And the
email file was lost in the hard-disk crash. So shut up. And anyway
nobody
wants to hear this stuff. So shut up. And then there are those who are
too
ignorant to understand your language. So shut up.
Even so, we are all still here.
I sin, am sinned against. There is still a mountain to climb.
>
> The crux of the thing is, no matter how fast the linear
> processor is - all positional values are pre-ordained (before
> move 1) by the programmer. These values do not change during
> the game (although value set b, then, set c may be invoked)
> and neither does the program learn anything.
>
> (This has also been a problem in a broader field of robotics.
> Advances were made by interacting with pyschologist, then with
> choleopterists.)
>
Chris Guinness is good for you.
Beetles. I'm talking beetles. How do they do it? If you'd
looked into U.Sussex like I said you would know this. The
roboticists are all over the biologist - a disgusting sight,
but its science folks!
One thing that I thought interesting a few months ago - is
that you and some other cove wandered off and had a good
conversation about how it all could be, and noted that humans
play chess differently than computers (humans are not only
linear, etc) then you both agreed to limit these possibilities
by waffling on your understanding of psychology. You set
limits and strange attributes which do not exist.
(fact-idiots)
Another pyschological aspect is the good old subject filtering
of the experiment. If you think this is a useful area to
explore - consult a real expert, on your own approach to what
*you* do, and also what is going on in the human player.
Good to have you back - I've saved all these nasty ideas for
you (and a few others) since you have some possibility of
doing something about them.
(Don't mention the war)
Red Three, over.
> Phil Innes
>
>
>
>Ed Schroder
>
Rolf Tueschen
<920805147.28471.0...@news.demon.co.uk>:
>I think you just misunderstood me.
>Now maybe I am misunderstanding you. What are you trying to tell me, you
>actually want to go and operate CSTal in Paderborn ? Or the created
>archetypes are real. Or what ?
I talk on several levels. It's an art like GM chess. Usually you don't
talk about -- yourself. Yes, in a way I said, if it's so, then send me.
But I see that you only wanted to gain cheap attention. In real you had
no question. So forget that level from my side.
Another level was so important. But also here you failed to understand.
Your problem, Sir.
If you'd asked me if Ed is more clever in reacting, I wouldn't say yes.
As I told you, it's an art. But nothing secret.
You are right in the other post. You guys rather hate each other. No
wonder that you can't react with your brain alone in computer chess.
As Ed said to me, in the end you become parano by force in thatt
business.
Chris Whittington
Robert Hyatt wrote in message <7bueif$a24$2...@juniper.cis.uab.edu>...
>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
>
>: Robert Hyatt wrote in message <7bu6bg$7ne$2...@juniper.cis.uab.edu>...
>:>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
>:>
>:>
>:>Then improve the program so that the 18-2 result doesn't happen,
>
>: Doesn't when we play single games.
>
>:> rather than
>:>complaining when it does and trying to justify why it happened. The game
>:>of chess is about _winning_ and _only about_ winning.
>
>: Bollocks. Bollocls. Bollocks.
>
>:> Last time I looked,
>:>no one could win a tournament by playing beautiful chess.
>
>: Could you tell ?
>
>: Chris Whittington
>
>
>Absolutely yes. And I'd rather win a game beautifully myself. But
>when it gets down to 'it', I'd rather win, period. IE the choice
>between playing a wild sac and going down in flames or slowly winning
>a KPP vs KP ending is not much of a choice, IMHO.
Right, let's go for the wild sac. That was what you meant wasn't it ?
Phil Innes
: This is the tail-end of a post by Ed Shroeder - to Chris W
: To compete in this area you have to make your program a weaker
: positional player????
: And the fact that important human-comp tournaments (AEGON,
: Harvard cup) are gone makes comp-comp even more important
: or even worse you can ask yourself the question: what else is left?
: Human comp tournaments played by normal chess rules would
: always result in the defeat of the computer by humans rated
: 2000. Without the opening book, is any program stronger than
: 1900, even running at 500Mhz?
Bob Hyatt wrote: Yes they are. _significantly_ stronger. The primary
advantage of the
book is to avoid repetitious behavior. But even that has been
eliminated
in at least a few programs. I don't think the book is a huge advantage
at
all, as I have seen lots of programs that would follow opening theory
without
any operational book at all. Many current programs _might_ not do this,
since it is easy for the author to rely on the book and ignore any sort
of
'opening evaluation'. But most worry about openings like 1. h3 and 2.
a3
and put some sort of developmental evaluation into their program to
avoid
getting killed when the opponent does that.
This is a crude argument on my part - but, with a disabled
opening book I can beat eveything, even blindfold with someone
calling out the moves. I am not an IM.
Another argument is that once I have won a Queens Indian
variation, or whatever, I always win it. (Its true that I
sometimes lose a game, but I have all the usual excuses, not
pysched up - playing at home distracted, etc. LOL!)
: And more important, what is the way to go?
: Learn why humans play (potentially) better chess. Emulate it.
: Get off your "we the programmers control the thing entirely"
: attitude and revisit the subject by inviting other opinion.
: The crux of the thing is, no matter how fast the linear
: processor is - all positional values are pre-ordained (before
: move 1) by the programmer. These values do not change during
: the game (although value set b, then, set c may be invoked)
: and neither does the program learn anything.
Not totally true... there are lots of 'second-order' evaluation things
done in programs... ie things that 'interrelate' like king exposure and
attracting pieces toward the king.
Yes - I should have made a more relative statement.
I agree that programs don't 'adjust' these weights based on game results
so that they 'learn' of course... But the set of 'weights' is so non-
linear that it would be very difficult for a human to 'use' this
information
in a match, for example. IE if you could detect the precise 'bishop
pair'
bonus, you might use this to exploit the program (Roman did this against
me
for a long time, continually complaining about 'it is too high', 'now it
is too
low'. It is a dynamic term (now) that includes two bishops, plus bishop
mobility, pawn structure, and so forth. Which means that in position A
it
might be value X, but in position B it will be different. So
exploitation is
not so easy...
I am sure I am being arrogant - but I do not need this
advantage. The engine cannot evaluate the current position
without reference to several sets of pre-programmed tables. As
a player I would maintain that the state-of-the-art is still
only 2000ish.
I grant you that in some positions it may be better - but not
overall. I think these advantages are only relevant one
program to another, and are not of the calibre that will
promote the worth of an engine 100+ points.
: (This has also been a problem in a broader field of robotics.
: Advances were made by interacting with pyschologist, then with
: choleopterists.)
To respond to my own point - how do beetles do it?
Why are they better at it than super-computers?
: Phil Innes
: Ed Schroder
Chris Whittington
Rolf Tueschen wrote in message <7buglm$q9s$1...@news02.btx.dtag.de>...
>"Chris Whittington" <ch...@cpsoft.demon.co.uk> wrote in
><920805147.28471.0...@news.demon.co.uk>:
>
>>I think you just misunderstood me.
>
>>Now maybe I am misunderstanding you. What are you trying to tell me, you
>>actually want to go and operate CSTal in Paderborn ? Or the created
>>archetypes are real. Or what ?
>
>I talk on several levels. It's an art like GM chess. Usually you don't
>talk about -- yourself. Yes, in a way I said, if it's so, then send me.
>But I see that you only wanted to gain cheap attention. In real you had
>no question. So forget that level from my side.
Cheap attention always. Also rich attention.
You want to go operate. My problem is that I don't want to pay $500 for
their pina coladas.
If you can raise subscriptions. And I'm sure the world would contribute just
to see you there, then fine. You go. I think the entire computer chess
community would just love it. Seriously. Rolf operating CSTal at Paderborn.
>
>Another level was so important. But also here you failed to understand.
>Your problem, Sir.
>
>If you'd asked me if Ed is more clever in reacting, I wouldn't say yes.
>
>As I told you, it's an art. But nothing secret.
Sorry, Pope. But if you speak to us you gotta come down to our level at
times.
>
>You are right in the other post. You guys rather hate each other. No
>wonder that you can't react with your brain alone in computer chess.
>
>As Ed said to me, in the end you become parano by force in thatt
>business.
>
It's not forced. The personalities together create instabilities that flip
too easily into war. But it doesn't have to be.
mclane
>>Your posts are abstract and are very coloured by emotions.
>Of course. I'm alive.
roman empire does not like this...
>>Not a good
>>combination; it's hard to extract any real information.
>Sorry, I'll try to be a database.
! :-))) 6.4
>I have lots of experience of this. Knowledge will always get better
>position. nps is fairly irrelevant for that. nps will, can, may still
>wriggle out. I think the critical comes at about 4 plies extra search.
>CSTal *is* the test for them. Its complex and its different. If they don't
>try to understand what's going on, they should quit reviewing.
>Chris Whittington
! :-)))
best wishes
Jesper Antonsson
>Jesper Antonsson wrote...
>>Your posts are abstract and are very coloured by emotions.
>Of course. I'm alive.
>
>>Not a good
>>combination; it's hard to extract any real information.
>
>Sorry, I'll try to be a database.
You could aim for making sense and stop your constant childish whining
about bean counters.
>We can discuss the feely sticks. Interesting idea. Mine sees round corners,
>and probes in more than one dimension.
>
>Feely sticks and minefields. You like the idea ?
Chess is a minefield, and I like that, yes. You use alpha-beta. You use
extensions. You have eval terms. Maybe you have slower eval terms or more
of them, maybe you opt for more nonlinearity, but as I see it, you really
don't do anything fundamentally different. You are welcome to prove me
wrong, of course.
>Suppose adding an eval term costs more time - so you throw it out right,
>pragmatic grounds ?
No. Throw it out if it don't win more games. More NPS is not automatically
a good thing, unless you get it without comprimising anything else, but
winning games *is* a good thing.
>But suppose the weight of ALL those eval terms, combined together yielded
>somehting really powerful. You'ld never know ?
As you add eval terms, your programs get worse, but some day the old terms
will get good because of new ones? Correct? You hope that some day
non-linearity will save your approach? That won't happen.
>If they took off exponential ? Only by BELIEVING can you find that.
>Pragmatism means no progress off the plateau.
Is that so? Your favourite bean-counter Crafty has been getting better
and better, and this while adding knowledge and slowing down it's search.
Where's the plateau there?
>>If your eval terms let you play a really good positional game, then it
>>won't fall for many pitfalls when playing other computers, rather it
>>would more often get good enough positions so that a faster searcher just
>>suddenly finds itself in a lost position and can't do anything about it.
>>Weren't this what Ed claimed happened in the one handicapped game against
>>Crafty?
>
>I have lots of experience of this. Knowledge will always get better
>position. nps is fairly irrelevant for that. nps will, can, may still
>wriggle out. I think the critical comes at about 4 plies extra search.
Ok, good. If you get this better position and win it more often than not,
you know you're on the right track. Until then, stop trying to put down
others that use a more successful approach than you. You only make
yourself look bad, not them.
>>And when Chris self-righteously talk about bean-counters, that others are
>>dumb, and that reviews of CSTal is the ultimate test for determining
>>the worth of the reviewing chess journalists, I frankly get a bit sick...
>
>Good. Take it all the way and vomit.
>
>CSTal *is* the test for them. Its complex and its different. If they don't
>try to understand what's going on, they should quit reviewing.
Whatever. Listen, it's ok for you to be proud of your program, even if too
much of it can get silly, but bashing others' is not. I thought you english
should be gentlemen...
Jesper
Robert Hyatt
: Right, let's go for the wild sac. That was what you meant wasn't it ?
: Chris Whittington
Not exactly. :) IE I am more interested in (a) 1-0 than in (b) 0-1
but white played a brilliant move but failed to follow up and lost.
Chris Whittington
Robert Hyatt wrote in message <7buppk$dqj$1...@juniper.cis.uab.edu>...
>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
>
>
>: Right, let's go for the wild sac. That was what you meant wasn't it ?
>
>
>: Chris Whittington
>
>Not exactly. :) IE I am more interested in (a) 1-0 than in (b) 0-1
>but white played a brilliant move but failed to follow up and lost.
Why ?
It's a game. Games are for fun and amusement and new ideas and learning
things. This is why young animals play games.
So why is 1-0 better ?
1-0 is single-minded. 1-0 is selfish. 1-0 is hate.
No ?
Chris Whittington
Robert Hyatt
I will gladly accept your challenge. :) In fact, if you'd like to
play a 6 game match vs crafty, I'm willing with one proviso: that each
time you play the same color you have to play a different opening. I
don't think it will 'wreck' its position at all, but I could be wrong.
But it would be an interesting experiment. I once played Joel Benjamin
a match with Cray Blitz, no book (couldn't get it copied to the machine
due to some network problems) and it did fine. Joel won the 'match'
because he played the same opening each time and would only vary when
the computer beat him...
So I believe computers can do quite well. Can you trick it with your
pet Latvian? Don't know. But I strongly doubt it as I think Crafty
will lean toward 1. d4 type openings left on its own, although this may
well have changed since I don't try this very often.
: Another argument is that once I have won a Queens Indian
: variation, or whatever, I always win it. (Its true that I
: sometimes lose a game, but I have all the usual excuses, not
: pysched up - playing at home distracted, etc. LOL!)
That may or may not work. Crafty has some 'learning' things that it
does, even with _no_ book, to provide some variability. IE that was done
in defense of those trying offbeat openings to take it out of book on
move 2.
: : And more important, what is the way to go?
This is possibly true... I have _never_ thought of chess programs as
"positional geniuses". However, the evaluation, even in a relatively
simple program such as crafty, is _very_ dynamic in nature. IE I'd bet
you could play 1,000 games against it and still not understand everything
it is evaluating. Because there are _so_ many ways to combine and
interact with those hundreds of scoring terms that I do use...
: I grant you that in some positions it may be better - but not
: overall. I think these advantages are only relevant one
: program to another, and are not of the calibre that will
: promote the worth of an engine 100+ points.
: : (This has also been a problem in a broader field of robotics.
: : Advances were made by interacting with pyschologist, then with
: : choleopterists.)
: To respond to my own point - how do beetles do it?
: Why are they better at it than super-computers?
: : Phil Innes
: : Ed Schroder
Robert Hyatt
: Robert Hyatt wrote in message <7buppk$dqj$1...@juniper.cis.uab.edu>...
:>Chris Whittington <ch...@cpsoft.demon.co.uk> wrote:
:>
:>
:>: Right, let's go for the wild sac. That was what you meant wasn't it ?
:>
:>
:>: Chris Whittington
:>
:>Not exactly. :) IE I am more interested in (a) 1-0 than in (b) 0-1
:>but white played a brilliant move but failed to follow up and lost.
: Why ?
: It's a game. Games are for fun and amusement and new ideas and learning
: things. This is why young animals play games.
: So why is 1-0 better ?
: 1-0 is single-minded. 1-0 is selfish. 1-0 is hate.
: No ?
Not in my opinion. I have played _lots_ of games over the years.
Football, basketball, tennis, chess, othello, you name it... and
in each case, the root issue was 'competition'. Had nothing to
do with 'selfish' or 'hate'... just about competing to prove you
are better than your opponent.
Yes, some games are not meant to be 'won'. IE I'd play Kasparov in
a heartbeat, just to play him, with very little hope of drawing or
winning. But I wouldn't do it day after day, unless I felt that it
was making me better, and that I would win _more_ as a result...
IE 'competition' is about winning and losing more than anything
else. Otherwise, why design a game where there must be a loser
for every winner? IE in tennis, just hit the ball back and forth
all afternoon? And it gets boring, because there is nothhing to
'drive your play upward'.
C. L. Williams
>Technically, it is all 0x88, my move generator makes 240,000 nps, bitmaps,
>what is a node and how do you count them, and 'I wrote some special code to
>help my program from playing Bxa7 - trapped bishop'.
>Utterly trivial. All they talk about is the mechanics of making a program,
>actually the mechanics of making a bean-counter program that does even know
>attacks-to, attacks-from or what is and isn't pinned, let alone whether
>anything can give check. They are stuck in the frigging coding. Making it
>faster. That's where the discussion level is. Thank you Hyatt, mostly. Thank
>you programmers of strength mostly 1700.
Where are all the posts on trivial programming techniques that are
preventing higher level concepts from being discussed? And even if
there were a lot of these bean-counter discussions by 1700 types, how
does it prevent other discussions from going on? Just curious.
Chuck
>Every time I see the word bitmap, I release the safety catch on my revolver.
>Frigging bitmaps. So someone thought a 64-bit register could map the 64
>square chess board. Wow ! Or Doh ! So incredibly trivial.
>Or somone thought that a rank, file or diagonal was, at most 8 long, so you
>could use 8-bit lookups to tell you stuff. Wow ! I think my friend Bart
>Simpson could have come up with that.
>Yet this talk of bitmaps is a serious claim to fame, no ? They rant on about
>it, like its the biggest deal of all time. When all it is, is a way to
>represent the data.
>What they then 'do' with this data is never discussed. How it relates to
>chess concepts is never discussed. And, the real design question for a chess
>program, whether they generate attacks-from, attacks-to and pin information,
>or whether they don't, or whether they half do it, or whether they might do
>it on the fly - such a primitive design decision, utterly integral to what
>type and style of program gets produced - again, never discussed. It's all
>just the one way of the Crafty this, Crafty that, and Crafty the other. All
>other ways of looking are spammed.
>The result, a mass of easy-to-build fast searchers, spamming everybody, and
>distorting the entire field.
>When Rolf says, he could talk and be useful, if the level was raised. It's
>true. If you got beyond bitmaps and basic wheels and cogs of the engine, and
>on to what to do with the data that's represented, what type of data should
>be represented, what type of data do you need to encapsulate dynamic
>knowledge, should a program know this, or that, or the other ? What new
>properties can emerge from different ways of searching, or different
>knowledge encodement ? All the next set of higher level questions. Then the
>people doing the engine design would actually be able to talk to those
>people who understood chess.
>But, of course, we know this is a pipe-dream. Firstly the programmers are
>too bogged down in implementations. And the chess knowledge people aren't
>very interested anyway. So forget it.
>Chris McGuinness is bad for you.
>>
>>Ed Schroder
>>
>>
>>>You make a review of a reviewer that is probably even more shallow,
>>>nonsensical and meaningless as the same reviewers jibing at me of 18-2
>last
>>>year in a pathetic attempt to try and get at me through my program.
>>>
>>>If he does that he'll write biased on anything. Does he like the
>programmer
>>>? Nice worded review. Does he not like the programmer ? Ignore it, or jibe
>>>18-2. This is the autoplayer overnight and go to sleep champion, no ?
>>>
>>>If Enrique really understood computer chess programs, or loved them, or
>>>wanted to describe the essential issues involved for 'his' readers, or
>>>whatever, he would have shown it with a senseful review of CSTal. He never
>>>has done - never even tried. If they can't write sensefully on this
>>>particular program, they're no good, they're stuck someplace. It is *the*
>>>test for them. They can't get away with the usual bollocks and
>platitudes -
>>>they have to step beyond. If they can't, well - the conclusion is obvious.
>>>
>>>
>>>Chris Whittington
>>>
>>>
>>>>
>>>>>>>I tell you what. Forget about all the tuning over-night with
>autoplayer.
>>>>>>>Simply forget it. Have the guts to simply deny to do it. Let the rest
>do
>>>>>>>it. And I for one promise you one thing. If you can show that your
>>>>>>>approach leads you to better results in games vs, humans you will be
>the
>>>>>>>new leader of the pack. The actual and past tradition of imbreeding in
>>>>>>>comp v comp is nuts Ed, completely inhuman nonsense. And the clients
>>>>>>>don't want the REBEL who beats them all (machine-wise) but they want a
>>>>>>>hell of machine that is able to play human beings. Totally different
>>>>>>>thing.
>>>>>
>>>>>>I of course agree.
>>>>>
>>>>>>Just one remark, there are a lot of computer chess enthusiastic who do
>>>>>>love comp-comp and quick solving difficult positions and I want to
>serve
>>>>>>them too.
>>
>>>>>
>>>>>Ed, do you want to make me immortal? :)
>>>>>
>>>>>I give you here as the only one the following idea under usual
>>>>>copyright. But you can use it if you want. Perhaps you know the
>>>>>ChessBase/Fritz/Mate engine concept? Just try to make an engine version
>>>>>of REBEL that does exactly what you mentioned. For the sake of fun,
>>>>>write a naked-hell-as-fast engine for autoplayers. Then write a version,
>>>>>just call it _R o l f i n g m o d e_ [with a tongue in cheek] where
>>>>>REBEL does search especially in complicated tactical positions as fast
>>>>>as hell again. Look at the 'Fritz correspondance analysis mode'. Just to
>>>>>understand my idea of several modes. The content is different of course.
>>>>
>>>>I leave that to Christophe.
>>>>
>>>>
>>>>>But to come back to our idea, your main playing REBEL should be the one
>>>>>tuned against human beings. I cannot tell you how happy you made me with
>>>>>your post.
>>>>>
>>>>>Now to your justified concern about those who like to analyse comp v
>>>>>comp. I wouldn't defame such a hobby. And in the beginning of all the cc
>>>>>it was the only thing you could do because comps/micros simply were too
>>>>>weak for in detail human play. But we have a different situation now. In
>>>>>the end most of the guys will understand that now it's time to
>>>>>reconsider what they are doing. But as I said above I would especially
>>>>>provide this group with an autoplayer version. Someone asked here if you
>>>>>wanted to die in beauty. I think you were the first who simply
>>>>
>>>>>understood the new times. We endusers want a strong "human" machine. BTW
>>>>>I have some ideas about the books. But I shouldn't talk here in public.
>>>>>Why giving away all new ideas?
>>>>
>>>>Again I entirely agree with you. Like to add that Chris was the first one
>>>>who described the problem in his own funny way.
>>>>
>>>>
>>>>>(Ed, I hope you get a first impression how stupid any exclusion even of
>>>>>the most stupid newbie might be, a fallacy so to speak. Human beings
>>>>>have very different strengths. What you techno beasts need is the help
>>>>>of people who are _not_ in the business but close enough to know what cc
>>>>>program is all about _and_ able to do so-called 'brainstorming' to
>>>>>provide you with "new" or notr so new ideas. Know what I mean? Let's
>>>>>start that. Life is too short ... I'm waiting with literally minimum 30
>>>>>new ideas a tempo. Now think of all the other members too. Do you really
>>>>>believe that burnt out older experts will have all the new ideas
>>>>>themselves?)
>>>>
>>>>When I look at my database there are over 100 suggestions to improve
>>>>my engine and I don't expect that this is different with others
>>>>
>>>>Ed Schroder
>>>>
>>>>
>>>
>>
Phil Innes
: opening book I can beat eveything, even blindfold with someone
: calling out the moves. I am not an IM.
I will gladly accept your challenge. :) In fact, if you'd like to
play a 6 game match vs crafty, I'm willing with one proviso: that each
time you play the same color you have to play a different opening.
LOL! That's a weakness you agree on then.
And I already play all sorts of programs -its not a challenge,
its a fact.
*****I will make a seperate post, however, to illustrate this
long-view thing. This is my main point, and we can see how
Kasparov preys on Karpov.
I don't think it will 'wreck' its position at all, but I could be
wrong.
But it would be an interesting experiment. I once played Joel Benjamin
a match with Cray Blitz, no book (couldn't get it copied to the machine
due to some network problems) and it did fine. Joel won the 'match'
because he played the same opening each time and would only vary when
the computer beat him...
I tried to interest Chris in a line - but (LOL!) he overrode
his computer and found an interesting move himself ...Ne4
(which loses Chris) then we never got to the long-ply event
that I wondered how CSTal would handle.
So I believe computers can do quite well. Can you trick it with your
pet Latvian? Don't know.
Not at all (besides this is not my variation, and someone may
be mighty teed off! ROFL - but we already tried to set this
up, you and I, ROFL) - I only used the Latvian as a stalking
horse in the conversation. I have been looking at some
Koltanowski recently - some extraordinarily complex positions
- Lasker reviews the variation and says, "it is completely
unclear who has an advantage, not for beginners." Which is
quite an admission for Lasker.
But I strongly doubt it as I think Crafty
will lean toward 1. d4 type openings left on its own, although this may
well have changed since I don't try this very often.
I promise to play only pawn sac variations, of the Mod.
Benoni, KID Saemisch, Blumenfeld or Benko.
As I mention above, even that very good analyst Lasker looks
at move 15 and throws his hands up (I can't believe he wrote
that in his book) saying "who knows" but these difficult
evaluations are usually because of a pawn-for-initiative
event, or a King-hunt, and the poor program has to try and run
between all the relative aspects of the potential of both
sides and the *absolute* factor of mate.
Phil Innes
preventing higher level concepts from being discussed? And even if
there were a lot of these bean-counter discussions by 1700 types, how
does it prevent other discussions from going on? Just curious.
Chuck
Because all the discussions add up to 10 points per year of
engine increase.
This fact is disguised by the fact that computer programs do
not play in regular tournaments, and have no real basis for
claiming any ratings that they have not earned.
Ratings of computers are mostly established when they play
each other, or at blitz or an inference from certain benchmark
performances.
Unfortunately, after 10,000 words have been exchanged on some
feature of a program, there may be an improvement in the chess
engine that amounts to a minute increase in its ability to
play chess.
Your questions helps illustrate that The Big Picture is about
the ability of chess engines to play chess, and Chris is
indicating (as I understand him) that we are getting to the
end of the road *in the ability of the engine to play chess*
by current methods.
People do not like this sort of post here. And there has been
a little tension around the subject. Cordially, Phil Innes
Phil Innes
Kasparov in their World Championship Match, Moscow 1985, Game 16.
Sicilian Defence, Taimanov Variation (!note!)
The preamble is that Kasparov had played it before and gave Karpov +
team a full week to analyse it. Considering that the line can be
considered unsound against such a formidable first tier player we get an
insight into gambling Garry, and also into Karpov.
What would be fascinating is how a computer reacts to the game - does it
see the bust by white? Does it see the very deep black plan, even
through a TN by white (never played before).
White Karpov : Black Kasparov
1. e4 c5
2. Nf3 e6
3. d4 cxd4
4. Nxd4 Nc6
5. Nb5 d6
6. c4 Nf6
7. N1c3 a6
8. Na3
pretty standard - a good analysis of the position is obtained in The
Sicilian Taimanov, by MarkTaimanov himself, and this much is in most
computer opening books.
8. ... d5 -the gambit called the Kasparov gambit for a while
9. cxd5 exd5
10. exd5 Nb4
11. Ne2 ! (exclam. by J. Nunn. And this is the TN by Karpov, as Nun says
Bc4 Bg4 has lead to short draws.
11... Bc5
but Nunn/Emms say that fantastical move has a flaw. It now interests me
(Phil) to know what various programs make of the position.
------
Kasparov got to move 19 with his teams' home preparation, including, I
emphasise, the TN 11. Ne2.
Now white can bust black's plan on move 12. I am sure your computer will
even prefer the bust, since it maintains the extra pawn and removes the
bind.
------
but Karpov played
12. 00
can your (anyones') program analyse who is better after 19. ....Bd6 and
how to get there?
Phil
Dann Corbit
news:7bvfm8$kjp$1...@juniper.cis.uab.edu...
[snip]
>Because it is part of 'the game'. Selfish generally means "keeping
>something to yourself'. I don't consider playing a 'game' to fit that
>category. Because both players agree to play, both have an expectation
>of wanting to win, etc...
>
>If the goal isn't 'to win' then there isn't much point in 'playing the
>game'.
>
>IMHO of course..
Unless the game has blunders, winning is never ugly. Now, 120 move
sister-kissers -- that's repugnant (unless you start as black, in which case
it is only mildly repulsive). I suppose some may liken it to a mammoth
soccer match that ends nil-nil (which I would also despise). Give me
victories and NBA action any day.
--
C-FAQ: http://www.eskimo.com/~scs/C-faq/top.html
"The C-FAQ Book" ISBN 0-201-84519-9
C.A.P. Newsgroup http://www.dejanews.com/~c_a_p
C.A.P. FAQ: ftp://38.168.214.175/pub/Chess%20Analysis%20Project%20FAQ.htm
Ed Schroder
>I think I leave it to a Dejas miner to demonstrate that this isn't so.
>
>Nevertheless, you were right in another post. The 'reviewer,
>so-called-experts' are a hangover from a few years ago when the programs
>were much weaker. Then the 'experts' could understand a bit and write some
>sense. But, strength-wise, the programs went right through and beyond them,
>so the 'experts' then resorted to autoplaying, sleep and 6.4. Put simply,
>program performance and properties are far too complex for them to
>comprehend anymore. So now reviews are full of 10.5-9.5 comp-comp, and
>platitudes such as 'human-like', or xyz is an 'interesting' program, or some
>other way of saying nothing in as long words as possible.
>
>Partly it is also our fault, because we never discuss anything. And, if we
>try, we get spammed wtih Crafty this and Crafty that and Crafty the other.
>No higher level concepts ever get discussed on the ngs.
>
>Technically, it is all 0x88, my move generator makes 240,000 nps, bitmaps,
>what is a node and how do you count them, and 'I wrote some special code to
>help my program from playing Bxa7 - trapped bishop'.
>
>Utterly trivial. All they talk about is the mechanics of making a program,
>actually the mechanics of making a bean-counter program that does even know
>attacks-to, attacks-from or what is and isn't pinned, let alone whether
>anything can give check. They are stuck in the frigging coding. Making it
>faster. That's where the discussion level is. Thank you Hyatt, mostly. Thank
>you programmers of strength mostly 1700.
>
I perfectly understand the point you are making. But chess programs are
mainly judged on comp-comp results so what can you expect from chess
programmers?
So the key of the chess programmer's behavior lies in the way how his
program is judged in the magazines and internet.
You are forced to program bit-maps Chris.
No chance huh? :)
Ed Schroder
Bas Hamstra
Extremely interesting, your dilemma. It also proves what Bob has said for
years that tuning against humans and against computers are two quite
different things. Still there are examples of programs that do very well
against both: Hiarcs, The King, Mchess, Rebel.
I don't understand why the "Fritz" approach was not very successfull for
years, while now suddenly it is...? Fritz used to be crushed by Genius,
Rebel, Hiarcs, Mchess. Still it was the fastest searcher.
Please Ed, don't let all programs be the same. Don't go for Fritz variant
no. 99. Invent something new and amazing. As I hope to do... Why not hook up
to ICC and go for the first place? Much more inspiring than only comp-comp,
many *quite* different games to watch.
Bas Hamstra,
Ed Schroder heeft geschreven in bericht <36E128CC...@rebel.nl>...
>>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>
>To compete in this area you have to make your program a weaker
>positional player????
>
>And the fact that important human-comp tournaments (AEGON,
>Harvard cup) are gone makes comp-comp even more important
>or even worse you can ask yourself the question: what else is left?
>
>Thinking all of this over the last months I came more or less to the
>conclusion that we have entered a new area, that adding new extra
>chess knowledge works out counter productive and that progress
>in computer chess will stop or already has been stopped as I can
>not believe that removing chess knowledge is progress.
>
>The fact that Program_X scores better against Program_Y doesn't
>(necessarily) mean that X is better than Y. X just scores better
>against Y and that's all. Maybe Y scores a lot better against humans
>than X and Aegon certainly has proven that.
>
>So where do we stand now in computer chess?
>
>And more important, what is the way to go?
>
>Ed Schroder
>
>
bruce moreland
<ch...@cpsoft.demon.co.uk> wrote:
>Thorsten has told time and time again what to look for in CSTal. Doesn't
>seem to make much difference. Actually the only person I noticed making
>informed and revealing comment is Mr Bruce, ages ago. But nobody noticed.
You and I have had numerous disagreements and personality clashes.
However, I appreciate the direction you are taking with CST. I enjoy
the wild sacrificial style, which is not always correct, but it is
amazing how *close* to correct it seems to be. The attacks I have
seen seem to have substance, and they do appear out of nowhere.
I think it is likely that your program entertains its customers much
more than other programs do, and it probably also teaches them things
they can't learn from other programs.
I don't know whether or not your approach will ever result in a score
in excess of 50% against everything else on even hardware, but if I
were in your shoes I wouldn't care that much, I would keep optimizing
for insanity and hoping that potential customers could take their eyes
off the Swedish list long enough to contemplate issues of style.
Is that what I said then or did I leave something out?
bruce
bruce moreland
I have considered these issues carefully. My own goal is to create
something that is interesting and strong, and I'm not going to gut my
program's style just to get an extra quarter ply to use against
computers.
I think that in the long term a good eval function is necessary to
avoid catastrophic mis-evaluation of crucial features, against both
humans and computers.
I avoid things like the Swedish list, in part because they've never
invited me to send a version, and in part because I don't want to get
sucked into that whole core-wars deal.
I would much rather be ranked on the FIDE list, which is where every
player really belongs.
In Paderborn, assuming I am selected to go, I will be competing with
something that will also play, with no modifications, against the
humans on ICC, which is how I've always done it.
bruce
bruce moreland
<ch...@cpsoft.demon.co.uk> wrote:
>I still maintain you didn't bite the bullet and encode any dynamic
>knowledge. It's interesting to watch you, Bob. Because once you were the
>bean-counter par excellence. Then you listened to all the attacks, and you
>started to move, but I think it was never 'all the way'. So now you inhabit
>a kind of limbo land, neither knowledged, nor proper bean counter. We need
>to shake you about a bit more :)
I think that what Bob is doing is a natural evolution of what Bob is
doing, given where he started and the environment he plays in.
Bob has always had an eval function at the tips, and it's always had
significant stuff in it.
bruce
j...@yktvmv.watson.ibm.com
on Sun, 07 Mar 1999 10:01:58 +0100,
Ed Schroder <in...@rebel.nl> writes:
>>From: en...@intercom.es
>
>Thanks that so many people reacted, I don't know were to start in
>replying. Sofar I have only seen postings who more or less agree
>with the very basic problem I have tried to explain. I would love to
>see the opinion of people who have opposite feelings, this for the
>value of the discussion.
>
>I can imagine people say, "Ed shut up, comp-comp is *the* way
>to measure progress and playing strength", and defend it.
>
>Anyone?
I won't argue comp-comp is the only way to measure progress
but it is a very important way.
One reason is practical. It is all very well to say that you
are optimizing for play against humans but unless you have human
grandmasters willing to play a few hundred games to test every change
you make you usually have no way of knowing whether a particular change
is good or bad. Just because your intuition says adding some piece of
knowledge is good doesn't make it so. It is relatively much easier to
test changes against strong computer opponents. And I expect most
improvements against computers will also be improvements against humans.
Btw you say you tune against humans, but rebel 8 is fairly
easy to beat with a little practice because its opening play is not
varied enough. Once you find a winning line you can play it over and
over again. Testing on chess servers such as ICC would quickly make
such problems apparent, why don't you do this?
Another reason is comp-comp results are objective while saying
one program plays "nice" chess but another program plays "ugly" chess
is very subjective. It is easy to lose objectivity concerning your
creative efforts so it is important to have an unbiased means of
evaluating their merits. It may be annoying when a change you had
great hopes for and spent a lot of effort on makes the program weaker
in comp-comp testing but it is likely that the change was a bad idea.
It is a lot easier to make progress when you can see how you are doing.
Another issue is bugs. I expect avoiding bugs is very
important for chess strength since it just takes one bad move to
lose a game. If comp-comp testing uncovers bugs this is certainly
desirable. Btw one would expect high knowledge programs to be more
complicated and buggier than low knowledge programs. One wonders
how much of the success of low knowledge programs is simply due to
a relative absence of bugs.
In my view program strength is best evaluated by results
achieved against strong competition. I think the best competition
currently readily available is provided by ICC and I personally
discount the strength of programs which do not participate. On a
related note I will never buy a program that does not have an
ICC interface allowing me to play it on a server. I expect this
is the future of competitive chess, strong programs playing each
other on the net.
James B. Shearer
Frji58
I agree with the people who want a windows interface for Rebel . I
particularly find it difficult to figure out how to do analysis.
I've only just got into computer chess. I must admit that even when I heard
from a salesman "Rebel (or MChess or Hiarcs) has a more human style" I'd think
to myself "Nice, but what about the latest SSDF list?" Therefore I'd make my
decision on what the "rating" is, then I'd look at the interface.
Now I'm a bit more educated and after looking at Rebel's excellent web site I
can see the value of owning Rebel. (which I do)
This might be a radical suggestion, but I believe if you want to get more
sales from the general public it might be worth changing the name of Rebel to
something more "chess-like", or at least without the possible negative
connotations of "rebel" - I can imagine parents would not like to encourage
their children to be "rebellious" and would see "Chessmaster" as a more
apropriate gift or learning tool's name. Chess Genius has a certain allure to
it's name, while HIARCS somehow attracts, like HAL in 2001 Space Oddesey. See
what I mean?
I don;t know anything about computer chess programming, but given the ever
increasing speed of processors, is it possible the brute force speed aproach
will eventually enable programs like Fritz to play more "human like" after a
deep search?
regards,
Michael Ginat
C. L. Williams
>Where are all the posts on trivial programming techniques that are
>preventing higher level concepts from being discussed? And even if
>there were a lot of these bean-counter discussions by 1700 types, how
>does it prevent other discussions from going on? Just curious.
>Chuck
> Because all the discussions add up to 10 points per year of
> engine increase.
> This fact is disguised by the fact that computer programs do
> not play in regular tournaments, and have no real basis for
> claiming any ratings that they have not earned.
> Ratings of computers are mostly established when they play
> each other, or at blitz or an inference from certain benchmark
> performances.
> Unfortunately, after 10,000 words have been exchanged on some
> feature of a program, there may be an improvement in the chess
> engine that amounts to a minute increase in its ability to
> play chess.
> Your questions helps illustrate that The Big Picture is about
> the ability of chess engines to play chess, and Chris is
> indicating (as I understand him) that we are getting to the
> end of the road *in the ability of the engine to play chess*
> by current methods.
> People do not like this sort of post here. And there has been
> a little tension around the subject. Cordially, Phil Innes
Then the best solution is to form rec.games.chess.computer.advanced.
Chuck
Nebulous Rex
<ch...@cpsoft.demon.co.uk> wrote:
>It's a game. Games are for fun and amusement and new ideas and learning
>things.
I agree that games can be used for those reasons. There are other
reasons to play games, too. A game is what the player makes of it,
after all. Some people throw horrible tantrums when they lose games,
thus signifying the game was more than mere fun and amusement to them.
I mean, if it's just fun and amusement, you're not really playing to
win or to lose, right? You're just playing because the game is fun
and amusing and you might get new ideas and learn things. How does
what the other guy thinks preclude this?
>So why is 1-0 better ?
It's whatever you think it is. If someone claims it's better, then
the game is probably a competition to him. That's his opinion.
You're still having fun and learning things, so who cares, right?
>1-0 is single-minded. 1-0 is selfish. 1-0 is hate.
Don't play with those "1-0" guys, then. I mean, why associate
yourself with single-minded, selfish hate-mongers?
C. L. Williams
>Phil Innes <in...@sover.net> wrote:
>>Where are all the posts on trivial programming techniques that are
>>preventing higher level concepts from being discussed? And even if
>>there were a lot of these bean-counter discussions by 1700 types, how
>>does it prevent other discussions from going on? Just curious.
>>Chuck
>> Because all the discussions add up to 10 points per year of
>> engine increase.
>> This fact is disguised by the fact that computer programs do
>> not play in regular tournaments, and have no real basis for
>> claiming any ratings that they have not earned.
>> Ratings of computers are mostly established when they play
>> each other, or at blitz or an inference from certain benchmark
>> performances.
>> Unfortunately, after 10,000 words have been exchanged on some
>> feature of a program, there may be an improvement in the chess
>> engine that amounts to a minute increase in its ability to
>> play chess.
>> Your questions helps illustrate that The Big Picture is about
>> the ability of chess engines to play chess, and Chris is
>> indicating (as I understand him) that we are getting to the
>> end of the road *in the ability of the engine to play chess*
>> by current methods.
>> People do not like this sort of post here. And there has been
>> a little tension around the subject. Cordially, Phil Innes
>Then the best solution is to form rec.games.chess.computer.advanced.
>Chuck
Phil,
I shouldn't have written the last post. For the record, if there is a
question I have concerning programming, I do research it through
Dejanews to see if it has been discussed before. Usually it has. If
not, it is nice to find somebody willing to answer my questions, even
if they are basic. Nobody is born knowing how to write chess
programs. Most of the bandwidth here is consumed by flame wars, not
computer chess on any level. I don't wish to fuel the flames.
I apologize for the earlier sarcasm.
Also cordially,
Chuck
Nebulous Rex
wrote:
>Not in my opinion. I have played _lots_ of games over the years.
>Football, basketball, tennis, chess, othello, you name it... and
>in each case, the root issue was 'competition'. Had nothing to
>do with 'selfish' or 'hate'... just about competing to prove you
>are better than your opponent.
Doing something to "prove you are better than your opponent" isn't
selfish? How so?
Robert Hyatt
: On 7 Mar 1999 22:02:51 GMT, Robert Hyatt <hy...@crafty.cis.uab.edu>
: wrote:
Because it is part of 'the game'. Selfish generally means "keeping
something to yourself'. I don't consider playing a 'game' to fit that
category. Because both players agree to play, both have an expectation
of wanting to win, etc...
If the goal isn't 'to win' then there isn't much point in 'playing the
game'.
IMHO of course..
--
Nebulous Rex
wrote:
> Your questions helps illustrate that The Big Picture is about
> the ability of chess engines to play chess, and Chris is
> indicating (as I understand him) that we are getting to the
> end of the road *in the ability of the engine to play chess*
> by current methods.
This is true. A hand-tuned assembly language implementation of a
bubble sort is still a bubble sort. It will still be outgunned by a
sloppy BASIC implementation of a quick sort.
Andrew Templeton
I like you idea of preferring an FIDE rating for your program. Would it not be possible
to use the online chess servers as a playing ground for all the programs and then picking
up their ratings at monthly intervals and reporting them, much like SSDF. This would give
us all program vs human ratings across a wide spectrum. There could be separate lists for
performance in blitz and long games which also take into account the hardware of the
various participating machines.
Actually, if I'm not mistaken, most of the chess servers maintain rating lists on a
continuous basis. So the rating Database already exists from which our new rating list can
be extracted, if anyone cares to do it.
Just a thought.
Best regards,
Andy Templeton
Robert Hyatt
: On Sun, 07 Mar 1999 20:49:23 -0400, Phil Innes <in...@sover.net>
: wrote:
Maybe not... IE try that bubble sort on a machine with 64 processors, and
then try that quicksort on the same machine. :)
quicksort has a problem in parallel architectures. The bubblesort has no
such shortcoming...
IE the computing platform may change what is 'best'...
Robert Hyatt
: Bob - here is this very complex scenario played between Karpov and
: White Karpov : Black Kasparov
I ran crafty at this point. It thinks black's position is much better.
-.75 in fact, with -=good for black. It likes Nc3, although it toyed with
Nf4 and Be3 at early plies. I don't see how white can 'save that pawn' in this
position.
: ------
: Kasparov got to move 19 with his teams' home preparation, including, I
: emphasise, the TN 11. Ne2.
: Now white can bust black's plan on move 12. I am sure your computer will
: even prefer the bust, since it maintains the extra pawn and removes the
: bind.
I don't see how to save the pawn, but I think something is missing, because...
(see below)
: ------
: but Karpov played
: 12. 00
This is not possible. Assuming you mean O-O, because white's bishop
has not yet moved. Did something get cut incorrectly?
: can your (anyones') program analyse who is better after 19. ....Bd6 and
: how to get there?
Obviously something went wrong, because I don't see any moves beyond move
12 in your post...
Bob
Ed Schroder
>It's very complicated. Of course you are right in that without sufficient
>chess strength no proper evaluation on a personal basis can be done. Still,
>even with a Grandmaster judging your program it will be complicated. Give a
>grandmaster copies of Rebel 8 and Rebel 9. Ask him to make a detailed
>comparison in playing strength and style. I have no doubt he will be very
>hard pressed to give you something conclusive. How does he find out if 9 is
>40-50 points stronger than 8?
You are right, it's unlikely he can.
But people want to hear an answer to the question "is it stronger?"
If grandmasters can't who is able?
So reviewers picked comp-comp as a reference and made it more
or less the truth.
Correct view?
>How do you do that with human players? How do
>you explain what it is that player A is doing that gives him a rating of 2440
>as opposed to player B with 2480? Does one write something like has a
>slightly better feel (it would be funny reading that a program has a 'feel')
>for positions, etc...? With no concrete examples to fall upon such as tests,
>positions, and so forth, all reviews will end up saying exactly that as it
>has become a guessing game.
Yes it is a guessing game and since 486/P90 times.
>The company claims the program is 40-50 points
>stronger so let's see if we can discover what it is that brought about this
>increase, if there is indeed an increase. I suspect that the ONLY way to get
>a proper review in this case is with the help of the programmer. Give the
>reviewers a press kit so to speak.
That's exactly what I did with Rebel7. I was asked by a magazine to
provide material for an upcoming review. Since I had a very good
relationship with the magazine I dared to face them with "How are
you going to judge that Rebel7 is stronger than Rebel6, it is above
your understanding these days with the current hardware". Next
I provided a long list of engine improvements with corresponding
positions. The material wasn't used in the review when published.
>Tell them what it is you have introduced
>and modified. Although this may produce some biased results, it will also
>help the reviewer know what to look for. I reviewed a French film called 'The
>Hairdresser's Wife' some years back. Naturally, I hardly 'needed' the press
>kit that came along in order to review the film. Yet without it my review
>would not have been as thorough for the kit revealed details on the origin of
>the project, the filming, etc...
You had more luck than I then... :-)
Ed Schroder
> Albert Silver
Ed Schroder
>Subject: Re: Response to Ed
>Hello Ed,
>
>I agree with the people who want a windows interface for Rebel . I
>particularly find it difficult to figure out how to do analysis.
Menu LEVEL, then pick "Analyze".
>I've only just got into computer chess. I must admit that even when I heard
>from a salesman "Rebel (or MChess or Hiarcs) has a more human style" I'd think
>to myself "Nice, but what about the latest SSDF list?" Therefore I'd make my
>decision on what the "rating" is, then I'd look at the interface.
The SSDF list will give you an impression about the playing strength
between computers. It will not tell you how good a program plays against
humans, how attractive a chess engine is, its style, its analysis qualities and
so on.
Next I wonder who is able to see the difference in playing strength between
an elo rated player of 2500 and 2540. Unless you consider this as non-trivial
pick Rebel, Mchess, Hiarcs or CS-Tal as your choice if you prefer a chess
program with a human style. My personal choice is Mchess because I like
its style but that's just me the others are great programs too.
>Now I'm a bit more educated and after looking at Rebel's excellent web site I
>can see the value of owning Rebel. (which I do)
>
>This might be a radical suggestion, but I believe if you want to get more
>sales from the general public it might be worth changing the name of Rebel to
>something more "chess-like", or at least without the possible negative
>connotations of "rebel" - I can imagine parents would not like to encourage
>their children to be "rebellious" and would see "Chessmaster" as a more
>apropriate gift or learning tool's name. Chess Genius has a certain allure to
>it's name, while HIARCS somehow attracts, like HAL in 2001 Space Oddesey. See
>what I mean?
I know what you mean, I was told the name Rebel especially in USA
sometimes is associated with the big war in the past.
>I don;t know anything about computer chess programming, but given the ever
>increasing speed of processors, is it possible the brute force speed aproach
>will eventually enable programs like Fritz to play more "human like" after a
>deep search?
Sometimes yes, mostly no.
Ed Schroder
>regards,
>
>Michael Ginat
Ed Schroder
>Hi Ed,
>
>Extremely interesting, your dilemma. It also proves what Bob has said for
>years that tuning against humans and against computers are two quite
>different things. Still there are examples of programs that do very well
>against both: Hiarcs, The King, Mchess, Rebel.
Sure.... the risen problem (actually I shouldn't have called it a problem)
doesn't came clear to me after an Eureka experience as it was more a
process of months of thinking based on years of investigations, looking
at comp-comp games.
Note that I do not have a dilemma. Maybe my writing style implied that.
Let me try to re-phrase: "I have reported a (new?!) fashion (trend)"
Actually the trend is not so new as it roots go back years ago but
its potential and effects were hardly discussed and not recognized
in the way I have tried to describe it now as I have called the header:
The end of computer chess progress?
Note the "?"......
Computer chess development of the last years:
. make the program as fast as possible (NPS)
. even remove existing chess knowledge to ensure the high NPS
. this in combination with a low branch factor (null-move)
. add only the very basic chess knowledge (piece-square tables etc)
. make the program aggressive to ensure tactics in games.
It's my opinion it all seem to work in the comp-comp area.
Now for comp-comp lovers this is great news but what about the people
who use a chess program for analysis and playing games?
10 days ago I released Rebel10.0c as an engine update for Rebel10.
In 10c I removed some chess knowledge which made Rebel 30% faster,
next this 30% gain was used to add tactics. As a result 10c is a better
comp-comp player now but a lower positional player than the original
Rebel10.
This is actually what is happening the last years and I want to report
it in all its details in the hope it will be understood.
There are advantages and disadvantages. The advantages are for the
comp-comp lovers. The disadvantages are for people who use a chess
program for analysis and study.
My personal opinion is that the disadvantages are bigger than the
advantages but that's just me. It's my wish the computer chess
press will pick up the subject, explain and judge.
>I don't understand why the "Fritz" approach was not very successfull for
>years, while now suddenly it is...? Fritz used to be crushed by Genius,
>Rebel, Hiarcs, Mchess. Still it was the fastest searcher.
Not quite so. Fritz4 was heavily under-rated on SSDF because Fritz4
was tested with a 128Kb hash table (or so). Next it had a horrible en-
passant bug which cost the program many losses. When the bug
was corrected the Fritz4 elo went up and up in every new list.
>Please Ed, don't let all programs be the same. Don't go for Fritz variant
>no. 99. Invent something new and amazing. As I hope to do... Why not hook up
>to ICC and go for the first place? Much more inspiring than only comp-comp,
>many *quite* different games to watch.
I agree with you, ICC is a chess world of its own. Wish my time was not
so limited because to understand ICC in all its details is a process of
months in my opinion.
Ed Schroder
Chris Whittington
bruce moreland wrote in message <36e411a2...@news.seanet.com>...
>On Sun, 7 Mar 1999 17:50:33 -0000, "Chris Whittington"
><ch...@cpsoft.demon.co.uk> wrote:
>
>>Thorsten has told time and time again what to look for in CSTal. Doesn't
>>seem to make much difference. Actually the only person I noticed making
>>informed and revealing comment is Mr Bruce, ages ago. But nobody noticed.
>
>You and I have had numerous disagreements and personality clashes.
I believe the difference is that I don't try and censor you, either by the
use of the expression "fuck-you", nor by running a police state.
>However, I appreciate the direction you are taking with CST. I enjoy
>the wild sacrificial style, which is not always correct, but it is
>amazing how *close* to correct it seems to be. The attacks I have
>seen seem to have substance, and they do appear out of nowhere.
>
>I think it is likely that your program entertains its customers much
>more than other programs do, and it probably also teaches them things
>they can't learn from other programs.
>
>I don't know whether or not your approach will ever result in a score
>in excess of 50% against everything else on even hardware, but if I
>were in your shoes I wouldn't care that much, I would keep optimizing
>for insanity and hoping that potential customers could take their eyes
>off the Swedish list long enough to contemplate issues of style.
>
>Is that what I said then or did I leave something out?
>
You left something out. Anyway, it was only the germ of an idea, and since
you said "you gave up talking to me on technical issues" I don't think I can
be that bothered to enlighten you.
Chris Whittington
>bruce
>
Chris Whittington
>Give the
>reviewers a press kit so to speak. Tell them what it is you have introduced
>and modified.Then they just make lists. And still don't understand. And the readers don't
understand the lists.
Perhaps, but basically there are two different reviews needed as there are two different groups of buyers. There are the new buyers who want to know everything there is to know about the program. It's strengths and weaknesses. And then there are the older buyers who want to know why they should upgrade version 1245.6 to 1245.7 for the special price of only $50, but only if they place their order in the next 2 weeks. They will be paying close attention to the explanation.
Albert Silver
Conventionally, they are persuaded to part with their money on the grounds that a great leap forward of between 50-150 ELO points has been achieved by the great wizard programmer.
Normally it hasn't, or it is down to hardware, or incestuous SSDF comp-comp testing - and we're about at the point now where the endusers have woken up. Also they realise that another 100 ELO points doesn't make any difference since they (endusers) are mostly relatively weak so they all get smashed anyway. And, they're beginning to realise that rating lists only reflect 1-0 multiplied up by lots of quantity, and this isn't really what it is all about.
I mean, if I could take a coin, and cleverly hack bits out of it, or add little weights, so that I'ld win, more often than not, coin flipping contests against other coin-programmers; and then we made huge coin-elo lists about who did the best coin hacking - would you be very interested ?
Of course, you wouldn't. And the reason you wouldn't is that the game would be too trivial, you understood it too easy and it didn't capture your imagination - even though the process of list creation, and elo's was just the same.
And the reason, allegedly why you are interested in the same process for computer chess, is that 'chess' interests you, it is hard to understand, it captures your imagination, right ?
But instead of being interested in the chess (we see almost no evidence of that in computer chess on the news groups) you are interested in the lists and the elo's and the 1-0's. Even to the point where the original purpose is entirely lost, and the list and elo's and fighting about what is best by proxy, become the process. You've fetishised lists and 1-0's, and forgotten the original purpose (that was to play the game of chess, btw). This is pretty much Thorsten's point when he starts ranting about choosing a girl via a vis rating lists and counting beans.
You've reduced the whole thing to coin-flipping contests. 1-0 is all that matters. This reflects on nothing more than the personalities and psychology of the persons involved. Basically, you are all, how shall I put this, arseholes. For only arseholes could have done it. Even our European hooligan class, those that go to football, even they wouldn't fetishise the football league tables to the extent of denying the football. Their games are great emotional events. Your games are played down serial cables while you sleep.
I shall use the recent words of that great self-orienteer Bruce Moreland. "Fuck you".
Chris Whittington
>Although this may produce some biased results, it will also
>help the reviewer know what to look for.
Thorsten has told time and time again what to look for in CSTal. Doesn't
seem to make much difference. Actually the only person I noticed making
informed and revealing comment is Mr Bruce, ages ago. But nobody noticed.
Maybe it's all too complex and difficult now. And they seek answers to the
wrong questions.Chris Whittington
> I reviewed a French film called 'The
>Hairdresser's Wife' some years back. Naturally, I hardly 'needed' the press
>kit that came along in order to review the film. Yet without it my review
>would not have been as thorough for the kit revealed details on the origin
of
>the project, the filming, etc...
>
> Albert Silver
>
>>
>> >they have tried in css, you know what happened to hiarcs ??
>>
>> I think the approach (is) was good, the contents were horrible I agree.
>>
>> Ed Schroder
>>
>>
>
>-----------== Posted via Deja News, The Discussion Network ==----------
>http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Chris Whittington
Jesper Antonsson wrote in message
<2A55E69DDDC71B28.43735D20...@library-proxy.airnews.ne
t>...
>Chris Whittington wrote:
>>Jesper Antonsson wrote...
>
>>>Your posts are abstract and are very coloured by emotions.
>>Of course. I'm alive.
>>
>>>Not a good
>>>combination; it's hard to extract any real information.
>>
>>Sorry, I'll try to be a database.
>
>You could aim for making sense and stop your constant childish whining
>about bean counters.
>
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
I must stop my constant childish whining about bean counters.
Signed: Bart Simpson
Chris Whittington
Ed Schroder wrote in message <36E3100E...@rebel.nl>...
>>From: "Chris Whittington" <ch...@cpsoft.demon.co.uk>
>
I've learnt to expect very little from chess programmers, Ed.
>
>So the key of the chess programmer's behavior lies in the way how his
>program is judged in the magazines and internet.
Really ? I thought the key to his behaviour was his general personality. To
1-0 or not to 1-0 that is the question, whether it is .....
Chris Whittington