NIMZO 3 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Komputer Korner
Bert recently sent me a copy of Nimzo 3. What an amazing program!!!
I am wondering why anybody who had bought Nimzo 3 has not defended
it and why haven't they spoke up about it? On preliminary tests,
I now rate Nimzo 3 as the best chess program ever. It rates either a
28 or a 29 on my list but I can't decide yet, as I haven't finished
testing. I will be writing an article on this program for the ICD
Computer Chess Reports. Yes folks, I have done Nimzo 3 an injustice only
because I had never seen the program and thus couldn't comment on it.
Nimzo 3 is what we serious chess studiers have been looking for all
along.
It is superior in all 3 aspects of computer chess. 1) Opening book
editor
playing engine and database capability. This program is so good I can't
believe it. Sorry Ed, your program is now relegated to 2nd place for
playing engine programs. Maybe Nimzo 3 is not quite as strong as
Rebel 8, but it is as strong as CM5000 on Eric Hallsworth's latest list.
The ability to actually add chess knowledge
to the program is an unbelievable feature. the opening book editor is
only surpassed by Bookup and the database facilities are top notch.
Okay I am saying too much, but hope to have a complete article on the
ICD CCR site soon. Join now for the many other reviews already listed
that have been written by myself and other experienced writers. CCR is
a non profit site and all membership fees go towards the cost of the
site.
The reviews include amazing in depth articles on Tascbase which was
the best all around program until I saw Nimzo 3.
LONG LIVE NIMZO. LONG LIVE NIMZO LONG LIVE NIMZO LONG LIVE NIMZO
--
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
and kouldn't find the real motive in chessbase.
ccmu...@aol.com
The new Version 3.5 (Jarkarta) is out now. And it is free for registered
users.
Its stronger and has some features more.
Its great.
Lonnie Cook
Yeap,ole Lonnie must just have to swallow his foolish pride to aquire this temptress of
Caissa.Forgive me (sob,sob,sob,sob, sniff,sniff,...)
Lonnie J. Cook
<lonni...@riconnect.com>
"Lonnie" on A-FICS,E-FICS,MMEICS & SICS
"SonsofThunder" on afics,efics,mmeics & sics
"DoctorWho" on ICC
Chris Whittington
me...@concentric.net wrote:
>
> On 1 Dec 1996 19:57:45 GMT, ccmu...@aol.com said something kinda
> like:
Unfortunately the main USA distributor turned it down.
Chris Whittington
>
> jeff
>
Chris Whittington
Komputer Korner <kor...@netcom.ca> wrote:
>
> Bert recently sent me a copy of Nimzo 3. What an amazing program!!!
> I am wondering why anybody who had bought Nimzo 3 has not defended
> it and why haven't they spoke up about it? On preliminary tests,
> I now rate Nimzo 3 as the best chess program ever. It rates either a
> 28 or a 29 on my list but I can't decide yet, as I haven't finished
> testing. I will be writing an article on this program for the ICD
> Computer Chess Reports. Yes folks, I have done Nimzo 3 an injustice only
> because I had never seen the program and thus couldn't comment on it.
> Nimzo 3 is what we serious chess studiers have been looking for all
> along.
> It is superior in all 3 aspects of computer chess. 1) Opening book
> editor
> playing engine and database capability. This program is so good I can't
> believe it. Sorry Ed, your program is now relegated to 2nd place for
> playing engine programs. Maybe Nimzo 3 is not quite as strong as
> Rebel 8, but it is as strong as CM5000 on Eric Hallsworth's latest list.
> The ability to actually add chess knowledge
> to the program is an unbelievable feature.
Before you get too carried away, you should realise, and I expect
you haven't, that Nimzo is a pre-processor.
That means that the board status is analysed before the search and
then piece-square tables are set up to guide the search.
Typically, a queen side pawn majority would promote knowledge which
gave a bonus for throwing forward the queen sides pawns, and then
(if appropriate) positioning pieces so that they attacked c6.
Ie reward a queen on a4, and rooks on the c-file etc.
I'm not knocking this as a concept, nor knocking Nimzo, which is
undoubtedly a strong program with a good SSDF result.
But, preprocessing has its problems. In the main it is that
as the position gets further away from the root, the assumptions
made about the status of the game get further and further away
from the ply 1 reality.
It can lead to situations where the queen is bonus-ed for being on a4
when the queen no longer exists, or there is a black pawn on b5,
or all the action is taking place on the other side of the board, or,
or, or.
The knowledge editor (known as CHE in Nimzo) imparts knowledge
in this manner.
It's not too difficult to see that as the program searches deeper,
either with more time allowance, or with hardware improvements,
that the problem gets more and more severe.
Such pre-processors gain relatively less and less strength with
speed than to tip-node evaluators.
That's why Crafty, (and Ferret, I think), as well as most
other programs evaluate at tip nodes.
Nimzo is similar in this way to Fritz, another pre-processor.
Chris Whittington
> the opening book editor is
> only surpassed by Bookup and the database facilities are top notch.
> Okay I am saying too much, but hope to have a complete article on the
> ICD CCR site soon. Join now for the many other reviews already listed
> that have been written by myself and other experienced writers. CCR is
> a non profit site and all membership fees go towards the cost of the
> site.
> The reviews include amazing in depth articles on Tascbase which was
> the best all around program until I saw Nimzo 3.
> LONG LIVE NIMZO. LONG LIVE NIMZO LONG LIVE NIMZO LONG LIVE NIMZO
>
>
Lonnie Cook
On Mon, 02 Dec 1996 03:48:14 GMT, me...@concentric.net wrote:
>On 1 Dec 1996 19:57:45 GMT, ccmu...@aol.com said something kinda
>like:
>
>>The new Version 3.5 (Jarkarta) is out now. And it is free for registered
>>users.
>>
>>Its stronger and has some features more.
>>
>>
>>Its great.
>How much ($) and what platform (DOS, Win95?)
>
>jeff
>
Jeff,
Gambit-Soft homepage is here <http://www.gambitsoft.com/gambit1e.htm> Go into the Chess
Programs section. Every 1.50 DM is = to $1 USA. You will find Nimzo3 there.
Robert Hyatt
Chris Whittington (chr...@cpsoft.demon.co.uk) wrote:
:
Right, for both Crafty and Ferret (and Zarkov and more I'm sure). I
did root-planning 20 years ago when depth was lucky to hit 4 pliess. It
worked o.k. In Cray Blitz we threw it out, as did Berliner in HiTech,
Thompson in Belle, and every other "strong" program I know. If you are
good with the extensions, but start of knowing the king is at g8 and is
exposed, by the time you reach the q-search, you are piling every piece
up on g7 or h7, while the king is over at a8 wondering what the crowd
on the other side of the board is looking at. Or a majority that's
dissipated by the time the search reaches the end of a line...
It's a good trick, but when you aspire to play at the top level, this seems
like the wrong approach. Drop down a class or two, and doing everything at
the root might be good enough...
Torstein Hall
Robert Hyatt <hy...@crafty.cis.uab.edu> skrev i artikkelen
<580601$8...@juniper.cis.uab.edu>...
It cant be all that bad, its at #9 with 2394 SSDF rating points, equall or
higher than many end node planning programs.
Torstein Hall
Chris Whittington
That's true. there's more than one way to skin a cat.
But I think the point remains, that this approach results in a program
that doesn't improve so well as the tip-noders with time allowance
and hardware improvements.
Chris Whittington
> Torstein Hall
Peter W. Gillgasch
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
[ bobbit ]
> Right, for both Crafty and Ferret (and Zarkov and more I'm sure). I
> did root-planning 20 years ago when depth was lucky to hit 4 pliess. It
> worked o.k. In Cray Blitz we threw it out, as did Berliner in HiTech,
> Thompson in Belle, and every other "strong" program I know. If you are
> good with the extensions, but start of knowing the king is at g8 and is
> exposed, by the time you reach the q-search, you are piling every piece
> up on g7 or h7, while the king is over at a8 wondering what the crowd
> on the other side of the board is looking at. Or a majority that's
> dissipated by the time the search reaches the end of a line...
>
> It's a good trick, but when you aspire to play at the top level, this seems
> like the wrong approach. Drop down a class or two, and doing everything at
> the root might be good enough...
Some comments. It is my favourite topic...
#1: It can't be *that* bad. Nimzo is doing it, Genius (yes !!!) is
doing it, Fritz is doing it. Hitech did it to a large extend. The
programs by Kittinger do it as well.
As with every method in computer chess there are weak and there are
strong programs which do this, so what is the "when you aspire..."
stuff about ? *You* have failed using this idea, so what ?
#2: Some positional aspects are very long term and are unlikely to be
changed during the search if certain preconditions hold.
Example: We know that one side has a white coloured bishop while
the other side has a dark coloured bishop. We know that we are
close to the endgame. On which files to you want your passed pawns ?
Another one. We have a large blocked pawn chain on the board
which cannot be broken by pawn advances at the endpoint files.
What is the point to gun your rooks onto the squares behind
that chain [from the point of view of the owner of the rook] ?
#3: This method decouples the search and the eval, hence it decouples
time critical code with knowledge code to a certain extend, but it
is important to know how the prescanning and the endpoint evaluation
interact.
#4: If you don't understand the interaction of endpoint and root node
eval then the program sucks, yes. And if you are not willing to
shuffle knowledge around between endpoints and the root (both
directions) based on "chess" AND "programming" issues, then the
whole approach is doomed to failure.
This "but the position changes during the search" is the usual excuse.
I heard it for years... Usually this argument serves the sole purpose
to kill the discussion off quickly instead of discussing the topic
seriously.
-- Peter
Komputer Korner
Chris Whittington wrote:
>
snipped
snipped
>
> > the opening book editor is
> > only surpassed by Bookup and the database facilities are top notch.
> > Okay I am saying too much, but hope to have a complete article on the
> > ICD CCR site soon. Join now for the many other reviews already listed
> > that have been written by myself and other experienced writers. CCR is
> > a non profit site and all membership fees go towards the cost of the
> > site.
> > The reviews include amazing in depth articles on Tascbase which was
> > the best all around program until I saw Nimzo 3.
> > LONG LIVE NIMZO. LONG LIVE NIMZO LONG LIVE NIMZO LONG LIVE NIMZO
Thanks for the information, but I am not worshipping Nimzo because of
it's chess strength even though at fast time controls it does
pretty well. The reason is because of all the other features.
It has everything except a chess tree,backsolving and (the ability to
drag
and drop because it is only a DOS program). And the functions it has are
all
top notch. I am amazed at this program.
I now worship at the GOD OF NIMZO!!!!!!!!!!!!!!!!!!!!!!!!!!!!
NIMZO....NIMZO.......NIMZO.......NIMZO.......NIMZO......NIMZO......
NIMZO......NIMZO........
ATTENTION ALL NETTERS: ALL POSTS FROM KOMPUTER KORNER SHOULD BE
IGNORED UNTIL THE TEMPORARY NIMZO FEVER SUBSIDES.
(Korner's keyboard is now in Nimzo macro mode)
NIMZO.......NIMZO.......NIMZO......NIMZO.....NIMZO......NIMZO.....
--
Komputer Korner
Robert Hyatt
Peter W. Gillgasch (gil...@ilk.de) wrote:
: Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
:
: [ bobbit ]
:
: > Right, for both Crafty and Ferret (and Zarkov and more I'm sure). I
: > did root-planning 20 years ago when depth was lucky to hit 4 pliess. It
: > worked o.k. In Cray Blitz we threw it out, as did Berliner in HiTech,
: > Thompson in Belle, and every other "strong" program I know. If you are
: > exposed, by the time you reach the q-search, you are piling every piece
: > up on g7 or h7, while the king is over at a8 wondering what the crowd
: > on the other side of the board is looking at. Or a majority that's
: > dissipated by the time the search reaches the end of a line...
: >
: > It's a good trick, but when you aspire to play at the top level, this seems
: > like the wrong approach. Drop down a class or two, and doing everything at
: > the root might be good enough...
:
: Some comments. It is my favourite topic...
:
: #1: It can't be *that* bad. Nimzo is doing it, Genius (yes !!!) is
: doing it, Fritz is doing it. Hitech did it to a large extend. The
: programs by Kittinger do it as well.
Note that I didn't say "bad", just "it seems like the wrong approach".
There's a difference... read further for why...
:
: As with every method in computer chess there are weak and there are
: strong programs which do this, so what is the "when you aspire..."
: stuff about ? *You* have failed using this idea, so what ?
*I* haven't failed, because *I* am not finished yet. The point is, and
always has been, that stuff done at the root is less accurate. You are
incorrect about HiTech. Re-read the various papers Hans and company wrote.
You'll find this: "as the program improved, we found more and more information
migrating from the piece/square tables into the recognizers which are used to
evaluate terminal position." That doesn't compute with your statement above.
As far as the "if you aspire" point above... *my* goal is to develop the
best chess program I know how to. It is certainly *my opinion* that tip
evaluation is better than root evaluation. I think logic would have to
agree.
The above "semi-quote" came from "Some Innovations in HiTech" in one of the
Advances in Computer Chess books..
BTW all of the above programs you mention don't evaluate at ply=1. Most
evaluate some things at the root, and then evaluate other things all along
the path as pieces are moved, scores are updated. Dave does this. I believe
Frans does this as well.
:
: #2: Some positional aspects are very long term and are unlikely to be
: changed during the search if certain preconditions hold.
:
: Example: We know that one side has a white coloured bishop while
: the other side has a dark coloured bishop. We know that we are
: close to the endgame. On which files to you want your passed pawns ?
:
: Another one. We have a large blocked pawn chain on the board
: which cannot be broken by pawn advances at the endpoint files.
: What is the point to gun your rooks onto the squares behind
: that chain [from the point of view of the owner of the rook] ?
none, but it's a bad example. Because there are many more positions
that arise where the pawns are blockaded some of the time, and not at
other times. If things change during the search, the root evaluator misses
it, if it doesn't the tip evaluator does extra work. But the tip evaluator
doesn't make a mistake in handling it...
:
: #3: This method decouples the search and the eval, hence it decouples
: time critical code with knowledge code to a certain extend, but it
: is important to know how the prescanning and the endpoint evaluation
: interact.
I don't disagree with you totally on this, because I do some things at
the root in Crafty at present, but this is slowly going away. For
example, look at pre-evaluate in the early 9.x codes, and in the present
11.x code. hardly anything is left. I'm still doing king centralization
with this, but that's got to go because sometimes the king doesn't belong
in the middle, and sometimes you can't tell where it belongs until after
doing a 20+ ply search...
Here's the easy question, however: If *speed* were unimportant, where
would you do all of the evaluation? I can't imagine any other answer than
at the endpoints, because that is *the* most accurate place to do it, when
you know what is where and what is what. Now, if that is true, then why
start moving things back to the root? The answer is "speed." And it
sure sounds like a compromise to me, trading accuracy for speed. no?
:
: #4: If you don't understand the interaction of endpoint and root node
: eval then the program sucks, yes. And if you are not willing to
: shuffle knowledge around between endpoints and the root (both
: directions) based on "chess" AND "programming" issues, then the
: whole approach is doomed to failure.
:
: This "but the position changes during the search" is the usual excuse.
: I heard it for years... Usually this argument serves the sole purpose
: to kill the discussion off quickly instead of discussing the topic
: seriously.
:
: -- Peter
However, it's *still* a valid point. Unless your root evaluation is good
enough to not make mistakes. Otherwise you could trash the search totally,
right? :)
Peter W. Gillgasch
I know that this is a touchy subject since it involves speed/accuracy
trade-offs and as I still was at the DarkThought group we had horrible
fights about that, so if anybody enters this thread *please* no flames
and no insider info :-) Seriously.
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
> Peter W. Gillgasch (gil...@ilk.de) wrote:
[ parity ply pruning applied ]
> : Some comments. It is my favourite topic...
> :
> : #1: It can't be *that* bad. Nimzo is doing it, Genius (yes !!!) is
> : doing it, Fritz is doing it. Hitech did it to a large extend. The
> : programs by Kittinger do it as well.
>
> Note that I didn't say "bad", just "it seems like the wrong approach".
> There's a difference... read further for why...
>
> :
> : As with every method in computer chess there are weak and there are
> : strong programs which do this, so what is the "when you aspire..."
> : stuff about ? *You* have failed using this idea, so what ?
>
> *I* haven't failed, because *I* am not finished yet. The point is, and
> always has been, that stuff done at the root is less accurate. You are
> incorrect about HiTech. Re-read the various papers Hans and company
> wrote. You'll find this: "as the program improved, we found more and more
> information migrating from the piece/square tables into the recognizers
> which are used to evaluate terminal position."
Yeah I know that. I never was interested in having no endpoint
evaluation. I am interested to have a strong hybrid program based on
Hans' statement "if you can do it in the PST, do it there". See, I can
throw "semi-quotes" around as well :-)
> That doesn't compute with your statement above.
> As far as the "if you aspire" point above... *my* goal is to develop the
> best chess program I know how to. It is certainly *my opinion* that tip
> evaluation is better than root evaluation. I think logic would have to
> agree.
Sure. If you don't factor in the following:
1. Speed / knowledge tradeoffs
2. Expandability
3. Complexity of knowledge and code
4. Dimishing returns for knowledge that can hurt you to the point where
overall performance takes a nose dive.
Of course there are dimishing returns for depth of search as well, but
IMHO a complex endpoint evaluator is just a vampire that sucks blood
from the speed of the underlying engine and is "the easy way out" that
finally leads to a total break down when you can't add anything without
loosing search depth. Then you need either a bigger / faster machine or
a new searcher and the whole process of writing a chess program starts
again.
Hence I like the more minimalistic approaches to computer chess.
> The above "semi-quote" came from "Some Innovations in HiTech" in one of
> the Advances in Computer Chess books..
But this doesn't tell the full story 8^)
> BTW all of the above programs you mention don't evaluate at ply=1. Most
> evaluate some things at the root, and then evaluate other things all along
> the path as pieces are moved, scores are updated. Dave does this. I
> believe Frans does this as well.
Yes. As I said, evaluating at ply 0 *alone* is of course nonsense. But
all of them do *significant* evaluation at ply 0, which is then
augmented by some small endpoint evaluation and/or incremental updating,
pretty much like Hitech did it.
> : #2: Some positional aspects are very long term and are unlikely to be
> : changed during the search if certain preconditions hold.
> :
> : Example: We know that one side has a white coloured bishop while the
> : other side has a dark coloured bishop. We know that we are close to
> : the endgame. On which files to you want your passed pawns ?
> :
> : Another one. We have a large blocked pawn chain on the board which
> : cannot be broken by pawn advances at the endpoint files. What is the
> : point to gun your rooks onto the squares behind that chain [from the
> : point of view of the owner of the rook] ?
>
> none, but it's a bad example. Because there are many more positions that
> arise where the pawns are blockaded some of the time, and not at other
> times. If things change during the search, the root evaluator misses it,
> if it doesn't the tip evaluator does extra work.
Ok, but as I said *if* strong preconditions hold it makes sense. If they
don't hold it doesn't. If they hold I make a "bet" on the relevant
search space (that is the space that is not in the "proof" part of the
tree) since I assume that certain structures can only be broken by
either a loosing sacrifice (bad eval, no point in evaluating too much)
or a combination that wins significant material (good eval, no point in
evaluating too much).
> But the tip evaluator doesn't make a mistake in handling it...
But is is outsearched, so it doesn't get to demonstrate this 8^)
DarkThought (the version that played at tourneys) was always an
"endpoint" evaluator and it got badly crushed by Frans two times. I am
no longer tolerating this :-)
> : #3: This method decouples the search and the eval, hence it decouples
> : time critical code with knowledge code to a certain extend, but it is
> : important to know how the prescanning and the endpoint evaluation
> : interact.
>
> I don't disagree with you totally on this, because I do some things at the
> root in Crafty at present, but this is slowly going away. For example,
> look at pre-evaluate in the early 9.x codes, and in the present 11.x code.
> hardly anything is left. I'm still doing king centralization with this,
> but that's got to go because sometimes the king doesn't belong in the
> middle, and sometimes you can't tell where it belongs until after doing a
> 20+ ply search...
And you can do things one could only dream about at the endpoints since
execution time is no issue, and you can even paractise good software
enginieering techniques without worrying about the overhead. You can
even supply an interface to the program that lets the user add knowledge
like Chrilly Donninger did with Nimzo. In other words you can build a
*system* instead of a monolithic "black box".
> Here's the easy question, however: If *speed* were unimportant, where
> would you do all of the evaluation? I can't imagine any other answer than
> at the endpoints, because that is *the* most accurate place to do it, when
> you know what is where and what is what.
Of course. But chess programming is to a large extend *programming* and
with the gap in processor speed and memory speed becoming wider each day
I'd rather have 1 mb of object code and tables to execute once to
provide the knowledge and a 8 k program running from on chip cache to
handle the search with a minimal eval that augments the oracle
knowledge.
Bob, you have a tremendous amount of experiece. If you have the option
to handcode a small program for a given machine and you intend to port
it around and to make changes over time, which one will you prefer:
(a) 8 k time critical searcher + 128 k "speed is no issue" 'C'
to evaluate with only the searcher "nailed onto the chip".
(b) 128 k time critical searcher + evaluation, everything losely
designed for a given architectural feature.
I'd take (a) since I am too lazy :-)
> Now, if that is true, then why
> start moving things back to the root? The answer is "speed." And it
> sure sounds like a compromise to me, trading accuracy for speed. no?
Yes. What is bad about that ? No Chris I don't want to listen 8^)
> :
> : #4: If you don't understand the interaction of endpoint and root node
> : eval then the program sucks, yes. And if you are not willing to shuffle
> : knowledge around between endpoints and the root (both directions) based
> : on "chess" AND "programming" issues, then the whole approach is doomed
> : to failure.
Note that I said "shuffle knowledge around" in both directions, so
obviously we are slightly talking about different things, I am talking
about "hybrid" programs with a strong "oracle" influence, you somehow
misunderstood this as "100% prescan or bust" mentality.
> : This "but the position changes during the search" is the usual excuse. I
> : heard it for years... Usually this argument serves the sole purpose to
> : kill the discussion off quickly instead of discussing the topic
> : seriously.
> :
> : -- Peter
>
> However, it's *still* a valid point. Unless your root evaluation is good
> enough to not make mistakes. Otherwise you could trash the search
> totally, right? :)
That is no serious argument since this is true for endpoint evaluations
as well. How many mistakes does your eval make,right ? :) Do you include
all those special cases at the endpoints ? No you don't. So you make
errors as well, your errors can easily be spotted since you simply need
to look at the current position and what the eval says, I need the
original root position as well as input, so things are (maybe) a bit
harder to fix, but speed is no issue, so the sky is the limit...
I don't want to convince anybody - I have given up on that some time ago
8^) - but I still don't see the point why many commercial programmers do
something and academic programmers believe it is a design failure.
Instead the academic programmers should think about ways how to overcome
the problems with that approach. Since this is clearly an "entry point"
for chess experts and/or database expertise I don't see why so little
work is done at this specific field, maybe because it is because of the
silent admission that evaluation is in any case error prone and mostly
trial and error, I don't know.
We are all "in the business" to produce highly tuned cute random number
generators for KK to put on his lists so let's don't fight about the
later statement 8^)
-- Peter
Chris Whittington
Get a life you mad git :)
Chris Whittington
Robert Hyatt
Right, but he did explain that not much could be done there...
:
: > That doesn't compute with your statement above.
I don't follow the above. in HiTech, *all* evaluation was done at endpoints
by the recognizer hardware, except for simple things left in the piece/square
tables. The main function of his "oracle" was to decide on the weights and
patterns to load into the recognizers. However, the hardware was all
"endpoint."
:
: > : #2: Some positional aspects are very long term and are unlikely to be
Here, Chris and I tend to converge. I accept the performance hit, because
as the depth increases I personally believe that a sophisticated eval will
improve quicker than a simple eval + slightly better tactics.
And here's the main point: If you could take all of crafty's endpoint
evaluation and move it to the root, so that evaluation took "0" time,
it would speed up some, but not much, because the eval code is about
30-35% of the total search time at present, and when (if) I ever get the
lazy eval to the state of what I did in Cray Blitz, this will likely drop
to 10%. That's not much of a performance hit at all... and given those
numbers, I'd stick where I am.
BTW, one other overlooked detail... as depth increases, the "incremental"
updaters actually slow down. Try fine #70... you go so deep, so fast,
that you are evaluating the king moves over and over and over, ending up
doing more work than if you only evaluated the king at the tips. at 12
plies + checks and captures, this is already not nearly as big a win as
it was at 5 plies... and it is getting worse as hardware speeds up...
:
: > :
I don't believe it is a "failure", but I *know* it is less reliable than
endpoint evaluation. Rather than tricking things out of the eval and back
to the root, I'm tricking things in the eval to minimize the penalty in
speed that full endpoint eval incurs.
I'm planning on taking the piece/square "summing" out of Evaluate() and
moving it into MakeMove(), but need to study how this behaves on deeper
searches. There's obviously a crossover point where it's worse than
endpoint eval, once the endgame is hit, but it gains some speed during
the middlegame. How much I'll find out, but notice I'm not taking anything
"global" out of the endpoint evaluation.. but without endpoint code, it's
impossible to see how pieces coordinate, and work together in a specific
position... and that I don't like.
:
: We are all "in the business" to produce highly tuned cute random number
Peter W. Gillgasch
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
[ lots of snipping ]
> Peter W. Gillgasch (gil...@ilk.de) wrote:
> : Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
> : > Peter W. Gillgasch (gil...@ilk.de) wrote:
> : Yeah I know that. I never was interested in having no endpoint
> : evaluation. I am interested to have a strong hybrid program based on
> : Hans' statement "if you can do it in the PST, do it there". See, I can
> : throw "semi-quotes" around as well :-)
>
> Right, but he did explain that not much could be done there...
Well if I recall correctly Hitech was a 2000-2050 program as it had only
the PST, then jumped by 200 as the recognizers were added and jumped
over 2400 after the Hitech group debugged things and gained experience.
The later numbers are FIDE performances in real "human" tourneys with
Hitech playing as citizien of Pennsylvania so if somebody sneers at that
and posts some SSDF numbers I'll flame him, this machine is horribly
outdated by now...
2000-2050 is not bad if you have "no eval at all" in a brandnew kid 8^)
But as I already said a "no eval at all" program is not the point of the
discussion.
> : Yes. As I said, evaluating at ply 0 *alone* is of course nonsense. But
> : all of them do *significant* evaluation at ply 0, which is then
> : augmented by some small endpoint evaluation and/or incremental updating,
> : pretty much like Hitech did it.
>
> I don't follow the above. in HiTech, *all* evaluation was done at endpoints
> by the recognizer hardware, except for simple things left in the piece/square
> tables. The main function of his "oracle" was to decide on the weights and
> patterns to load into the recognizers. However, the hardware was all
> "endpoint."
True, but the decision what patterns should be detected were done at the
root and the weights assigned to patterns were computed at the root as
well. Now who says you can't do that in software ? As I said analyzing
the root is *not* the same as having no endpoint eval at all. It is
simply tuning the eval to the current situation of the game, nothing
more, but nothing less.
So the problem is only partially "pure root node eval", it is also a
"what is important because we have *that* position on the board"
problem.
[ snip ]
> : > Now, if that is true, then why
> : > start moving things back to the root? The answer is "speed." And it
> : > sure sounds like a compromise to me, trading accuracy for speed. no?
> :
> : Yes. What is bad about that ? No Chris I don't want to listen 8^)
>
> Here, Chris and I tend to converge. I accept the performance hit, because
> as the depth increases I personally believe that a sophisticated eval will
> improve quicker than a simple eval + slightly better tactics.
If the tactics are only slightly better you are correct. If the tactics
are *significantly* better (say you are outsearched by 2 - 3 plies until
maybe the ending) then things change dramatically... If it turns out
that the eval of the "oracle" program is better for the current position
(you only need to make a single good move, you know, all in a row, but
only a single move, maybe for the wrong reason, it doesn't matter at
all) since it is more "high level" and detects things a "tip node"
evaluator doesn't grasp because of the need for speed (don't deny that
you often said "no I don't add this. Slows things down, never happens
anyway") then I'd say that the combination of search and knowledge is
lethal.
> And here's the main point: If you could take all of crafty's endpoint
> evaluation and move it to the root, so that evaluation took "0" time,
> it would speed up some, but not much, because the eval code is about
> 30-35% of the total search time at present, and when (if) I ever get the
> lazy eval to the state of what I did in Cray Blitz, this will likely drop
> to 10%. That's not much of a performance hit at all... and given those
> numbers, I'd stick where I am.
In your case this is of course a valid point. In my case I'd loose much
more of the speed...
Seriously, cache effects are weird. Unless you have a program that runs
basically completely "on chip" you don't know what weird trade-offs
apply. E.g. the rehashing you did in CB using 8 bits as offset... You
had the vector loads then, so it was ok. Did you change that on the
PeeCee to a simple linear scan to max out cache bandwidth ? If I don't
do that this costs me 40% on the PowerPC which is *both* arms and *both*
legs for me... If you put this in into Crafty for testing and it costs
you only 5% or something then you will understand that 35% for a full
blown eval is probably not the number I'd get...
> BTW, one other overlooked detail... as depth increases, the "incremental"
> updaters actually slow down. Try fine #70... you go so deep, so fast,
> that you are evaluating the king moves over and over and over, ending up
> doing more work than if you only evaluated the king at the tips. at 12
> plies + checks and captures, this is already not nearly as big a win as
> it was at 5 plies... and it is getting worse as hardware speeds up...
Now this is a very good and interesting point. I can't argue this,
beside the argument that the extra speed will prevent a pawn ending in
an inferior position to happen...
[ snip ]
> I don't believe it is a "failure", but I *know* it is less reliable than
> endpoint evaluation. Rather than tricking things out of the eval and back
> to the root, I'm tricking things in the eval to minimize the penalty in
> speed that full endpoint eval incurs.
Now that is the aspect I don't like. Tricking the eval code. Eval code
is never static, it changes over time and if you need to "trick" things
there, then you loose flexibility. This is the "software engineering"
aspect I was talking about. If I trick things in time critical code I
want to trick things that are *guaranteed* to improve performance, not
just in some isolated cases which may contribute nothing to the outcome
of a game. Throwing tons of well structured code at the problem is
another issue since it spoils your programming habits...
> I'm planning on taking the piece/square "summing" out of Evaluate() and
> moving it into MakeMove(), but need to study how this behaves on deeper
> searches.
Now hold it. You do that in eval() ? Why ? I mean the main advantage is
doing things in an incremental fashion. Apart from pathological cases it
should *always* be a win, if it isn't I'd say that something is grossly
wrong with Crafty...
> There's obviously a crossover point where it's worse than
> endpoint eval, once the endgame is hit, but it gains some speed during
> the middlegame. How much I'll find out, but notice I'm not taking anything
> "global" out of the endpoint evaluation.. but without endpoint code, it's
> impossible to see how pieces coordinate, and work together in a specific
> position... and that I don't like.
True, this problem isn't easy to solve, but I am convinced that it is
doable for 95% of the interesting cases at little cost... As I said the
PST is not the only thing one can have 8^)
-- Peter
Robert Hyatt
:
: > Peter W. Gillgasch (gil...@ilk.de) wrote:
: > : Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
: > : > Peter W. Gillgasch (gil...@ilk.de) wrote:
:
: > : Yeah I know that. I never was interested in having no endpoint
: > : evaluation. I am interested to have a strong hybrid program based on
: > : Hans' statement "if you can do it in the PST, do it there". See, I can
: > : throw "semi-quotes" around as well :-)
: >
: > Right, but he did explain that not much could be done there...
:
: Well if I recall correctly Hitech was a 2000-2050 program as it had only
: the PST, then jumped by 200 as the recognizers were added and jumped
: over 2400 after the Hitech group debugged things and gained experience.
: The later numbers are FIDE performances in real "human" tourneys with
: Hitech playing as citizien of Pennsylvania so if somebody sneers at that
: and posts some SSDF numbers I'll flame him, this machine is horribly
: outdated by now...
:
: 2000-2050 is not bad if you have "no eval at all" in a brandnew kid 8^)
:
: But as I already said a "no eval at all" program is not the point of the
: discussion.
2000-2050 is also about what would be expected from a program that searched to
a depth of 9+ plies in middlegame (no null move on the first version I don't
believe)... regardless of how sloppy the eval was about recognizing things.
And HiTech's not "horribly outdated"... It's still faster than pretty much
any program around, running right around the speed of the P6/200, and it has
some pretty good evaluation code on top of it. Dangerous is the word I'd
use, not "outdated." Belle is still *dangerous*... if it's still running
that is.
:
: > : Yes. As I said, evaluating at ply 0 *alone* is of course nonsense. But
: > : all of them do *significant* evaluation at ply 0, which is then
: > : augmented by some small endpoint evaluation and/or incremental updating,
: > : pretty much like Hitech did it.
: >
: > I don't follow the above. in HiTech, *all* evaluation was done at endpoints
: > by the recognizer hardware, except for simple things left in the piece/square
: > tables. The main function of his "oracle" was to decide on the weights and
: > patterns to load into the recognizers. However, the hardware was all
: > "endpoint."
:
: True, but the decision what patterns should be detected were done at the
: root and the weights assigned to patterns were computed at the root as
: well. Now who says you can't do that in software ? As I said analyzing
: the root is *not* the same as having no endpoint eval at all. It is
: simply tuning the eval to the current situation of the game, nothing
: more, but nothing less.
but their reasons were different. They only had hardware to analyze "x"
patterns at the tip, so they tried to intelligently decide which "X" patterns
would be most useful. In software this isn't much of an issue, other than in
terms of "cycle costs"...
:
: So the problem is only partially "pure root node eval", it is also a
: "what is important because we have *that* position on the board"
: problem.
:
: [ snip ]
:
: > : > Now, if that is true, then why
: > : > start moving things back to the root? The answer is "speed." And it
: > : > sure sounds like a compromise to me, trading accuracy for speed. no?
: > :
: > : Yes. What is bad about that ? No Chris I don't want to listen 8^)
: >
: > Here, Chris and I tend to converge. I accept the performance hit, because
: > as the depth increases I personally believe that a sophisticated eval will
: > improve quicker than a simple eval + slightly better tactics.
:
: If the tactics are only slightly better you are correct. If the tactics
: are *significantly* better (say you are outsearched by 2 - 3 plies until
: maybe the ending) then things change dramatically... If it turns out
: that the eval of the "oracle" program is better for the current position
: (you only need to make a single good move, you know, all in a row, but
: only a single move, maybe for the wrong reason, it doesn't matter at
: all) since it is more "high level" and detects things a "tip node"
: evaluator doesn't grasp because of the need for speed (don't deny that
: you often said "no I don't add this. Slows things down, never happens
: anyway") then I'd say that the combination of search and knowledge is
: lethal.
While I won't argue, you aren't going to get 2-3 plies by doing this
stuff. If you could somehow produce a hardware endpoint evaluator that
I could plug into Crafty, which would evaluate a position in under one
picosecond, you'd effectively drive the evaluation time to "0" and you
would speed Crafty up by 30%. That won't get a ply, much less 2-3...
Also evaluation done at the root, is done looking at a position that
might have nothing to do with the position you reach at the root. For
example an ending that starts off being KPP vs KP, and ends up being KQ vs KP
for example. A lot changed from the root to the tip, and the two positions
have nothing at all in common.
If you think that doesn't happen, I see similar things all the time. A position
with 14 pieces on the board, Crafty does a 13 ply search, and announces a mate in
40... by finding a way to force the trade of nearly everything to hit the endgame
database. going from 14 to 4 in one search is a huge gap from root to tip, and
it would be quite easy to screw up badly...
:
: > And here's the main point: If you could take all of crafty's endpoint
:
Here's what I've "done" (as opposed to what I'm "doing" now):
1. if things are absolutely static, I "root eval'ed them". For example,
piece centralization, since the center of the board is fixed, hopefully...
2. if things are dynamic, but change very rarely, then I incrementally
update them. examples might include pawn structure things like open files,
most advanced pawn on each file, least advanced pawn on each file, things
I used in the eval in CB, but which hardly ever changed during the search.
3. everything else is done at the tips. I don't take any chances with
"target squares" that are based on root analysis, because that target
could have run away by the time the tip node is reached.
In CB, I did 2 for speed only, so it wasn't done at every tip (3) position,
which saved time. However, I didn't do any "2" stuff for pieces, since they
are the subject of most moves in the tree.. In Crafty, I simply haven't
gotten around to (a) the lazy eval we used in CB, which will probably drive
the eval time down to around 10% or so, and (b) the incremental updates for
things that hardly ever change, but which take time to "notice" at tip
positions. Once I get this learning stuff put away, I plan on cleaning this
up quite a bit...
: [ snip ]
:
: > I don't believe it is a "failure", but I *know* it is less reliable than
: > endpoint evaluation. Rather than tricking things out of the eval and back
: > to the root, I'm tricking things in the eval to minimize the penalty in
: > speed that full endpoint eval incurs.
:
: Now that is the aspect I don't like. Tricking the eval code. Eval code
: is never static, it changes over time and if you need to "trick" things
: there, then you loose flexibility. This is the "software engineering"
: aspect I was talking about. If I trick things in time critical code I
: want to trick things that are *guaranteed* to improve performance, not
: just in some isolated cases which may contribute nothing to the outcome
: of a game. Throwing tons of well structured code at the problem is
: another issue since it spoils your programming habits...
By "trick" I mean "efficient programming". If you look at Crafty's eval,
there's no short-cuts at all as of yet. No "assumptions" either... It
simply analyzes what it thinks is appropriate. The bitmap evaluation offers
lots of room for optimizations that the standard offset approach doesn't of
course, but that's all the "trickiness" I do there...
:
: > I'm planning on taking the piece/square "summing" out of Evaluate() and
: > moving it into MakeMove(), but need to study how this behaves on deeper
: > searches.
:
: Now hold it. You do that in eval() ? Why ? I mean the main advantage is
: doing things in an incremental fashion. Apart from pathological cases it
: should *always* be a win, if it isn't I'd say that something is grossly
: wrong with Crafty...
Think about this: you have 8 pieces on the board. You search to depth
20. You do 20 piece/square updates incrementally, because every time you make
and unmake a move you have to update the piece/square value for that piece
on its new square. At tip only evaluation, you only do 8 piece/square
updates. 3X faster, *not* slower...
In the middlegame it's somewhat different of course, but it's not clear at
all that it wins anything... Time will tell...
:
: > There's obviously a crossover point where it's worse than
:
Chris Whittington
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
> I'm planning on taking the piece/square "summing" out of Evaluate() and
> moving it into MakeMove(), but need to study how this behaves on deeper
> searches. There's obviously a crossover point where it's worse than
> endpoint eval, once the endgame is hit, but it gains some speed during
> the middlegame. How much I'll find out, but notice I'm not taking anything
> "global" out of the endpoint evaluation..
>
> but without endpoint code, it's
> impossible to see how pieces coordinate, and work together in a specific
> position... and that I don't like.
This is of critical importance.
With tip-node evaluations some of the dynamic interactions between
pieces can be assessed. I don't see this happening at all via the
piece-square paradigm.
Chris Whittington
Robert Hyatt
Chris Whittington (chr...@cpsoft.demon.co.uk) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: >
:
: > I'm planning on taking the piece/square "summing" out of Evaluate() and
: > moving it into MakeMove(), but need to study how this behaves on deeper
: > searches. There's obviously a crossover point where it's worse than
: > endpoint eval, once the endgame is hit, but it gains some speed during
: > the middlegame. How much I'll find out, but notice I'm not taking anything
: > "global" out of the endpoint evaluation..
: >
: > but without endpoint code, it's
: > impossible to see how pieces coordinate, and work together in a specific
: > position... and that I don't like.
:
:
: With tip-node evaluations some of the dynamic interactions between
: pieces can be assessed. I don't see this happening at all via the
: piece-square paradigm.
:
: Chris Whittington
:
:
Right, unless you make your incremental code so complicated it becomes
slower than tip-node evaluation. One example is a bonus for doubling
rooks or a rook and queen on the 7th rank. If you move one rook, you
affect the score for the other since the doubling bonus it should have
is "gone." This can be accounted for incrementally, but suddenly you
do a lot of work for a single piece...
brucemo
> > moving it into MakeMove(), but need to study how this behaves on deeper
> > searches. There's obviously a crossover point where it's worse than
> > endpoint eval, once the endgame is hit, but it gains some speed during
> > the middlegame. How much I'll find out, but notice I'm not taking anything
> > "global" out of the endpoint evaluation..
> >
> > but without endpoint code, it's
> > impossible to see how pieces coordinate, and work together in a specific
> > position... and that I don't like.
>
>
> With tip-node evaluations some of the dynamic interactions between
> pieces can be assessed. I don't see this happening at all via the
> piece-square paradigm.
>
> Chris Whittington
Definitions:
First-order term: Something that is either hashed (like pawn structure or king
safety), or is computed extremely quickly, such as a piece-square value.
Second-order term: Something that requires you to do more complicated analysis of
the board position, at every node. A good example would be a mobility term
calculated during Eval by following vectors for all of the pieces.
I do the piece-square summing as part of MakeMove(), since I use the move-table
move generation paradigm I have the positional delta of the move sitting there in
the move table element. Perhaps it would be OK to sum in Eval, but having the
info earlier makes it easier to do lazy eval, and also keep me from having to
iterate through the pawns, at all, since I have no other second-order terms for
those (everything is in pawn structure eval).
You can still do some piece interaction stuff without second-order eval, you can
reward certain piece combinations such as the bishop pair without having to make
it a real second-order term, you just recompute when a capture is made, and have
this value accessible to the eval at some later point.
If you are talking about "doubled rooks on the 7th", yes, that sounds more
second-order, but if you just had a few terms like this you could probably figure
out a way to make them first order.
Even a program that evaluates at the tips can benefit from computing some stuff
higher in the tree.
bruce
Seifriz
me...@concentric.net wrote:
>
> On 1 Dec 1996 19:57:45 GMT, ccmu...@aol.com said something kinda
> like:
>
> >The new Version 3.5 (Jarkarta) is out now. And it is free for registered
> >users.
> >
> >Its stronger and has some features more.
> >
> >
> >Its great.
> How much ($) and what platform (DOS, Win95?)
>
> jeff
NIMZO 3.5 can be downloaded free from our site
http://www.gambitsoft.com
But it only runs if you have NIMZO 3, otherwise the upgrade
does not work! Nimzo is a DOS program, but it runs under
WIN 95 on our machine without problems!
TO KOMPUTER KORNER: tried to send email from 2 different
providers, it keeps coming back: return to sender, address
unknown, no such number on the phone...???
Bert!
Harald Faber
Hello chris,
I think I have an interesting aspect to tell you:
CW> > >The new Version 3.5 (Jarkarta) is out now. And it is free for
CW> > >registered users.
CW> > >Its stronger and has some features more.
Available on http://www.gambitsoft.com
CW> > >Its great.
CW> > How much ($) and what platform (DOS, Win95?)
DOS
Costs 198DM, divide it by 1.5 and you have US-$
CW> Unfortunately the main USA distributor turned it down.
You may try it at Steinwender EDV, I recently posted the adress concerning
the FRUT-utility.
Ciao and see ya
Harald
--
Peter W. Gillgasch
brucemo <bru...@nwlink.com> wrote:
> Robert Hyatt wrote:
>
> > *I* haven't failed, because *I* am not finished yet.
>
> This line deserves its own response, but what can I possibly say?
Since chess programmers are never finished they never fail.
Or something like that 8^)
-- Peter
p.s.: Bob, I am working on replies to the original discussion. But I am
horribly busy right now, so if I don't react soon that doesn't
mean that I haven't failed, because I am not finished yet 8^)
Gee. I realize that you created a kool :) one liner...
mclane
Komputer Korner <kor...@netcom.ca> wrote:
>Bert recently sent me a copy of Nimzo 3. What an amazing program!!!
>I am wondering why anybody who had bought Nimzo 3 has not defended
>it and why haven't they spoke up about it? On preliminary tests,
>I now rate Nimzo 3 as the best chess program ever.
ARGHHHHHHHHHHHHH !!!
How can you come to such an idiotic conclusion ???!!!
It has the knowledge in the first-ply of the search !!!!!!
So, you are fascinated by a program that knows about chess only in the
first ply !!!!!!
How can you really believe that NIMZO is the strongest program ?!
Strongest WHERE ?
In canada ???
On the philipines ?
In austria maybe. But austria is not the rest of the world.
>It rates either a
>28 or a 29 on my list but I can't decide yet, as I haven't finished
>testing. I will be writing an article on this program for the ICD
>Computer Chess Reports. Yes folks, I have done Nimzo 3 an injustice only
>because I had never seen the program and thus couldn't comment on it.
>Nimzo 3 is what we serious chess studiers have been looking for all
>along.
I thing you disqualify yourself with these statements.
>It is superior in all 3 aspects of computer chess. 1) Opening book
>editor
>playing engine and database capability. This program is so good I can't
>believe it.
Are you drunken ??
What is the superiority of the chess engine Nimzo has?!
Show us games or positions that show the superiority of this
pre-processing fast/stupid-software from austria.
>Sorry Ed, your program is now relegated to 2nd place for
>playing engine programs. Maybe Nimzo 3 is not quite as strong as
>Rebel 8, but it is as strong as CM5000 on Eric Hallsworth's latest list.
Haehhhhhh ???
Is it strong or not?
CM5000 ? Hallsworth ?
>The ability to actually add chess knowledge
>to the program is an unbelievable feature.
My dog can bark. This shows it is more intelligent!
AGAIN for the KOMPUTER KORNER's:
the knowledge is used in the first ply, not any further.
> the opening book editor is
>only surpassed by Bookup and the database facilities are top notch.
>Okay I am saying too much, but hope to have a complete article on the
>ICD CCR site soon.
>Join now for the many other reviews already listed
>that have been written by myself and other experienced writers.
It always wonders me how modest you are!!
YOU and the other experienced writers...
>CCR is
>a non profit site and all membership fees go towards the cost of the
>site.
>The reviews include amazing in depth articles on Tascbase which was
>the best all around program until I saw Nimzo 3.
>LONG LIVE NIMZO. LONG LIVE NIMZO LONG LIVE NIMZO LONG LIVE NIMZO
Some things die earlier than you will believe. Some things are even
dead...
>--
mclane
gambi...@t-online.de (Seifriz) wrote:
>me...@concentric.net wrote:
>>
>> On 1 Dec 1996 19:57:45 GMT, ccmu...@aol.com said something kinda
>> like:
>>
>> >The new Version 3.5 (Jarkarta) is out now. And it is free for registered
>> >users.
>> >
>> >Its stronger and has some features more.
>> >
>> >
>> >Its great.
>> How much ($) and what platform (DOS, Win95?)
>>
>> jeff
>NIMZO 3.5 can be downloaded free from our site
>http://www.gambitsoft.com
>But it only runs if you have NIMZO 3, otherwise the upgrade
>does not work! Nimzo is a DOS program, but it runs under
>WIN 95 on our machine without problems!
>TO KOMPUTER KORNER: tried to send email from 2 different
>providers, it keeps coming back: return to sender, address
>unknown, no such number on the phone...???
>Bert!
No such zone if I remeber elvis accurate.
Komputer Korner
I have 2 email numbers, but the old one is still good.
kor...@netcom.ca
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
kouldn't find the real motive in ChessBase and missed
the real learning feature of Nimzo. Long live Nimzo!!!!!!
brucemo
Robert Hyatt wrote:
> Right, unless you make your incremental code so complicated it becomes
> slower than tip-node evaluation. One example is a bonus for doubling
> rooks or a rook and queen on the 7th rank. If you move one rook, you
> affect the score for the other since the doubling bonus it should have
> is "gone." This can be accounted for incrementally, but suddenly you
> do a lot of work for a single piece...
This is a tangent, but you may be able to remove this term, since 2x the
bonus for a single rook on the 7th is still a lot, and there are often so
many horrendous tactics once you get two rooks on the 7th that even a
shallow search might gleefully report on the subsequent mayhem, so
perhaps this can be a "synthetic" term.
Same for Q+R on 7th. I don't have a doubled 7th-rank rook term, and no
term at all for queen on the 7th. Perhaps a search will prove
otherwise, but I bet you don't find many games between your program and
my program where this mattered.
bruce
brucemo
Robert Hyatt wrote:
> Think about this: you have 8 pieces on the board. You search to depth
> 20. You do 20 piece/square updates incrementally, because every time you make
> and unmake a move you have to update the piece/square value for that piece
> on its new square. At tip only evaluation, you only do 8 piece/square
> updates. 3X faster, *not* slower...
I think I detect a logic error.
Assume you have 8 pieces on the board, and for every position you evaluate, you have to
make and unmake every move in a 20-ply variation. This is a lot of adding and
subtracting if you are doing your piece-square crud in make and unmake.
But you don't have to make and unmake every move in a 20-ply variation every time you do
an eval. Think about it. If the move at the root is 1. e4, there are a heck of a lot of
calls to eval before you end up unmaking this move. So the cost of the incremental
piece-square update is shared amongst a heck of a lot of nodes. I have no numbers but
I'm guessing that you share so many of them amongst so many calls to eval, that only the
last few in the variation have any effect. If you've got these numbers, just compare the
number of makes and unmakes you do, with the number of evals you do multiplied by the
average number of pieces on the board. I bet the second number is way bigger in most
typical cases.
Maybe you can find a case wherein it isn't bigger, for instance low-material endings, but
the ending is typically the last part of any given game, and you've got a heck of a lot
of evals with more material on the board to do before that.
bruce
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
No, but I did find several games where the two rooks got to the 7th, but
crafty had something to offset this like one rook on the 7th plus an
outpost knight. The two rooks simply tie everything down to defending
the pawns/king. Hard to say qualitatively that it is better with it than
without, but this came about in the early development of Crafty because it
was getting into dead lost positions without realizing it. BTW, this bonus
is almost worth a whole pawn...
Bob
Chris Whittington
brucemo <bru...@nwlink.com> wrote:
>
> Robert Hyatt wrote:
>
> > Right, unless you make your incremental code so complicated it becomes
> > slower than tip-node evaluation. One example is a bonus for doubling
> > rooks or a rook and queen on the 7th rank. If you move one rook, you
> > affect the score for the other since the doubling bonus it should have
> > is "gone." This can be accounted for incrementally, but suddenly you
> > do a lot of work for a single piece...
>
> This is a tangent, but you may be able to remove this term, since 2x the
> bonus for a single rook on the 7th is still a lot, and there are often so
> many horrendous tactics once you get two rooks on the 7th that even a
> shallow search might gleefully report on the subsequent mayhem, so
> perhaps this can be a "synthetic" term.
>
> Same for Q+R on 7th. I don't have a doubled 7th-rank rook term, and no
> term at all for queen on the 7th. Perhaps a search will prove
> otherwise, but I bet you don't find many games between your program and
> my program where this mattered.
I've got a very manic term for two 7th rooks (it does try to
look for anomalies like king not on 8th and defended pawns and so
on), which has given some neat wins. But, on the other side, its
also caused some losses.
Chris Whittington
>
> bruce
A.Mader
Chris Whittington wrote:
:
: >
: > Bert recently sent me a copy of Nimzo 3. What an amazing program!!!
: > I am wondering why anybody who had bought Nimzo 3 has not defended
: > it and why haven't they spoke up about it? On preliminary tests,
: > 28 or a 29 on my list but I can't decide yet, as I haven't finished
: > testing. I will be writing an article on this program for the ICD
: > Computer Chess Reports. Yes folks, I have done Nimzo 3 an injustice
only
: > because I had never seen the program and thus couldn't comment on
it.
: > Nimzo 3 is what we serious chess studiers have been looking for all
: > along.
: > editor
: > playing engine and database capability. This program is so good I
can't
: > playing engine programs. Maybe Nimzo 3 is not quite as strong as
: > Rebel 8, but it is as strong as CM5000 on Eric Hallsworth's latest
list.
: > The ability to actually add chess knowledge
: > to the program is an unbelievable feature.
: Before you get too carried away, you should realise, and I expect
: you haven't, that Nimzo is a pre-processor.
:
: That means that the board status is analysed before the search and
: then piece-square tables are set up to guide the search.
:
: Typically, a queen side pawn majority would promote knowledge which
: gave a bonus for throwing forward the queen sides pawns, and then
: (if appropriate) positioning pieces so that they attacked c6.
:
: Ie reward a queen on a4, and rooks on the c-file etc.
:
: I'm not knocking this as a concept, nor knocking Nimzo, which is
: undoubtedly a strong program with a good SSDF result.
:
: But, preprocessing has its problems. In the main it is that
: as the position gets further away from the root, the assumptions
: made about the status of the game get further and further away
: from the ply 1 reality.
:
: It can lead to situations where the queen is bonus-ed for being on a4
: when the queen no longer exists, or there is a black pawn on b5,
: or, or.
I know that your program is no brute forcer and that it uses most of its
knowledge in the whole search tree. Chrilly has tried to implement this,
too. The result was that Nimzo played bad moves just because the good
move was in the tree! We have new ideas to handle this and maybe they
are implemented in Nimzo4. Didn't you have the same problem?
Your example: I don't know exactly what you mean. If the queen no longer
exists, there will be no bonus for a queen on a4. How good do you know
CHE? It is a kind of "IF pattern THEN advice" language. If the action is
taking place on the other side of the board, the pattern that leads to
the advice "Move Q to a4" will most probably not be on the board.
However, it is an "advice", not an "order"!
:
: The knowledge editor (known as CHE in Nimzo) imparts knowledge
: in this manner.
:
: It's not too difficult to see that as the program searches deeper,
: either with more time allowance, or with hardware improvements,
: that the problem gets more and more severe.
:
: Such pre-processors gain relatively less and less strength with
: speed than to tip-node evaluators.
That's not true in all cases! One good thing to use CHE is the early
middle game. When a program is out of book, it often plays bad moves
just because it doesn't understand the idea of the opening. CHE is a
_very_ good way to implement such things! In this part of the game there
are often many moves with nearly the same evaluation. With CHE, you can
help Nimzo with the decision! When I played against Nimzo and used my
special tournament opening I won most of the times. So I wrote a CHE
file giving Nimzo the ideas of the opening. Since then I am losing game
after game....
:
: That's why Crafty, (and Ferret, I think), as well as most
: other programs evaluate at tip nodes.
:
: Nimzo is similar in this way to Fritz, another pre-processor.
:
: Chris Whittington
:
Nothing wrong about Fritz, nothing wrong about Nimzo, nothing wrong
about CST! :)
Best wishes
Andreas Mader
: > the opening book editor is
: > only surpassed by Bookup and the database facilities are top notch.
: > Okay I am saying too much, but hope to have a complete article on
the
: > ICD CCR site soon. Join now for the many other reviews already
listed
: > that have been written by myself and other experienced writers. CCR
is
: > a non profit site and all membership fees go towards the cost of the
: > site.
: > The reviews include amazing in depth articles on Tascbase which was
: > the best all around program until I saw Nimzo 3.
: > LONG LIVE NIMZO. LONG LIVE NIMZO LONG LIVE NIMZO LONG LIVE NIMZO
: >
: >
: > --
: > The komputer that kouldn't keep a password safe from
: > prying eyes, kouldn't kompute the square root of 36^n,
: > and kouldn't find the real motive in chessbase.
Chris Whittington
Can you try and rephrase what you mean here, because I'm
lost to understand ...... :)
> We have new ideas to handle this and maybe they
> are implemented in Nimzo4. Didn't you have the same problem?
>
> Your example: I don't know exactly what you mean. If the queen no longer
> exists, there will be no bonus for a queen on a4. How good do you know
> CHE? It is a kind of "IF pattern THEN advice" language. If the action is
> taking place on the other side of the board, the pattern that leads to
> the advice "Move Q to a4" will most probably not be on the board.
> However, it is an "advice", not an "order"!
OK, rephrase. Suppose you target a Q+B+N attack on h7, with
the idea Qxh7 check and nasty follow ups.
So you increase weightings for Q + B + N hitting h7, and the Queen
gets exchanged. Still the B + N weights for the attack on h7 exist,
but for no good reason anymore ...... ?
>
> :
> : The knowledge editor (known as CHE in Nimzo) imparts knowledge
> : in this manner.
> :
> : It's not too difficult to see that as the program searches deeper,
> : either with more time allowance, or with hardware improvements,
> : that the problem gets more and more severe.
> :
> : Such pre-processors gain relatively less and less strength with
> : speed than to tip-node evaluators.
>
> That's not true in all cases! One good thing to use CHE is the early
> middle game. When a program is out of book, it often plays bad moves
> just because it doesn't understand the idea of the opening. CHE is a
> _very_ good way to implement such things! In this part of the game there
> are often many moves with nearly the same evaluation. With CHE, you can
> help Nimzo with the decision! When I played against Nimzo and used my
> special tournament opening I won most of the times. So I wrote a CHE
> file giving Nimzo the ideas of the opening. Since then I am losing game
> after game....
Fair enough. But you must recognise that as the search space
gets deeper the leaf node positions get further and further away
from the ply 1 reality ?
I was trying to make the point that this approach has a kind of
build in redundancy (like car manufacture) such that the program
does not maintain its improvement with larger search space as
leaf-node processors ...... ?
>
> :
> : That's why Crafty, (and Ferret, I think), as well as most
> : other programs evaluate at tip nodes.
> :
> : Nimzo is similar in this way to Fritz, another pre-processor.
> :
> : Chris Whittington
> :
>
> Nothing wrong about Fritz, nothing wrong about Nimzo, nothing wrong
> about CST! :)
Agreed. there's more than one way to skin a cat.
Alternatively there's something wrong with all of them, otherwise
Kasparov would have fallen :)
Chris Whittington
Peter Fendrich
A.Mader wrote:
(snip)
> Your example: I don't know exactly what you mean. If the queen no longer
> exists, there will be no bonus for a queen on a4. How good do you know
> CHE? It is a kind of "IF pattern THEN advice" language. If the action is
> taking place on the other side of the board, the pattern that leads to
> the advice "Move Q to a4" will most probably not be on the board.
> However, it is an "advice", not an "order"!
An idea for the NIMZO team:
- Add some kind of strength figure to the rules.
- Build a learning feature of the same type that Bob has done in
Crafty.
- Apply learning to the rule strength, instead of only the book moves.
This would be a really intresting project to follow!
The next step is to even invent new rules during learning. Something
completely different but I have some ideas for anyone interested.
(snip)
--
J-P Fendrich
mclane
"A.Mader" <andrea...@siemens.at> wrote:
The queen is on a4 in the first play of main-line.
It gets a bonus from your:
If queen is on a 4 then give move a bonus of 0.02 points in move list
order.
But the program moves the queen away in the main line, but THERE your
if cases are not looked after.
In the games between Nimzo 3.5 Jakarta and another program I am
testing in the moment, I have often seen that NIMZO
a) did not expect the moves that came
b) was instantly shocked when the move comes
c) gets from +0,30 into -1.40 after seconds although it was in the 9th
ply before.
In my opinion the reason could be the following:
a) much pruning (horizont)
b) wrong evaluation
c) no balance between having evaluation in first ply and NO evals in
further plies.
This causes horrible games, where I have the feeling I see how an ape
plays against a human.
I would say: Nimzo is the austrian version of Fritz.
It has more knowledge like Fritz, but it has the same
system-immanent-problems:
It does not understand the moves of the opponent, it has no plans
(despite of some tactical reasons) and if it has some plans, they are
only immanent in the first ply, often it places the pieces back in the
rest of the main-line. This is always one step forward and 3 steps
backwards.
Sometimes it will find a way, or if the opponent is weaker it will win
because of opponents fault, but this is no chess, this is ape-chess.
Randomnly played chess!
Chess by drunken apes!!
Maybe it is really Chrilly's cat, that is the mind behind the moves of
Nimzo now.
Also I have read Chrilly article in Computer-Schach-Spiele about you
and the Nimzo-team developing NIMZO.
The author writes that one has to play 5000 games to KNOW about a
program. Because he is statistician HE KNOWS this.
In my opinion only a fool or somebody who cannot play chess has to
play 5000 games to find out about a program.
But on the other hand I do know understand why the NIMZO team
has this big problems in findung out WHICH version is the stronger
one.
If you have to play 5000 games to get an impression, OH MY GOD!!!
I don't have to Kiss/F-ck/met 5000 girls to find out which ones I
like/love!
But you are statisticians. You know about error-margins (while
kissing?!) and you KNOW that almost 5000 games are NOT ENOUGH!
I think - hm - I don't want to be impolite again....
the author's of these strange statements should be tested if he's
drunken or if he had smoked cannabis (or is not a human being but a
cat=pet).
>:
>: The knowledge editor (known as CHE in Nimzo) imparts knowledge
>: in this manner.
>:
>: It's not too difficult to see that as the program searches deeper,
>: either with more time allowance, or with hardware improvements,
>: that the problem gets more and more severe.
>:
>: Such pre-processors gain relatively less and less strength with
>: speed than to tip-node evaluators.
Nimzo evaluated the position before the Kill-Move of the opponet came
with his PRE-PROCESSED evaluation as good and was happy with +0,30 and
some nice dreams about this and that move.
The next move was the 2nd move of the main-line for Nimzo on move
before. So it had not evaluated the ply much. It expected another
move, although this move was a kill-move. So it has pruned away the
move because of the WRONG evaluation in the first ply.
Then suddenly the Kill-move came and it instantly saw the mess because
NOW it evaluated the position AFTER the move was made (it was not
abnle to prune it away anymore, also it had an evaluation in this
position-level).
But now it cannot do anything further. The position is lost.
The opponent has seen that, and not only that: the opponent has
PLANNED the whole stuff, because only if a program is able to produce
the positions it likes, it can win.
Programs like fritz or nimzo are unable to produce anything they like.
They forget everything they liked after one ply, and then anything is
pure randomnly-outsearched without any sense or plan.
>That's not true in all cases! One good thing to use CHE is the early
>middle game. When a program is out of book, it often plays bad moves
>just because it doesn't understand the idea of the opening. CHE is a
>_very_ good way to implement such things! In this part of the game there
>are often many moves with nearly the same evaluation. With CHE, you can
>help Nimzo with the decision! When I played against Nimzo and used my
>special tournament opening I won most of the times. So I wrote a CHE
>file giving Nimzo the ideas of the opening. Since then I am losing game
>after game....
You would never try to teach your son this way!
Thats the first thing we teach a youngster, that BEFORE he moves the
move on the board he likes, he should ask himself:
what can my opponent play when I move this move!
Nimzo plays like a youngster that likes some moves, and instantly
moves them, and then seeing the mess.
Only sometimes search can stop him for instantly commiting suicide,
often it needs 3-7 moves after the opneing boook, normally.
>:
>: That's why Crafty, (and Ferret, I think), as well as most
>: other programs evaluate at tip nodes.
>:
>: Nimzo is similar in this way to Fritz, another pre-processor.
>:
Right: Nimzo is the austrian version of Fritz's
Same fame- same stupidity.
>: Chris Whittington
>:
>Nothing wrong about Fritz, nothing wrong about Nimzo, nothing wrong
>about CST! :)
I subscibe the same above sentence! Nothing against Nimzo or Fritz.
I have also bought them.
BTW: IN Computerschach und Spiele
DIETER STEINWENDER SELLS Nimzo 3.5 Jakarta for 39,- DM !!!!!!
On the server from GAMBIT SOFT I can download the same stuff for
NOTHING!
SO: is the Distributor of Nimzo = Dieter Steinwender really objective
when he publishes NICE articles about NIMZO when he WANTS to make much
money out of it ?
And it seems he wants to make moneay because he sells something that I
can get for free somewhere else.
A.Mader
Peter Fendrich wrote:
: An idea for the NIMZO team:
: - Add some kind of strength figure to the rules.
: - Build a learning feature of the same type that Bob has done in
: Crafty.
: - Apply learning to the rule strength, instead of only the book
moves.
:
: This would be a really intresting project to follow!
: The next step is to even invent new rules during learning. Something
: completely different but I have some ideas for anyone interested.
:
: (snip)
:
: --
: J-P Fendrich
Thank you for your ideas. I wrote about 20 pages full with my own ideas
a few months ago and gave it all to Chrilly. Maybe we will see some of
them in Nimzo4. I will speak with him about your suggestions, your first
one I offered him, too!
Best wishes
Andreas
A.Mader
Chris Whittington wrote:
> > I know that your program is no brute forcer and that it uses most of its
> > knowledge in the whole search tree. Chrilly has tried to implement this,
> > too. The result was that Nimzo played bad moves just because the good
> > move was in the tree!
>
> Can you try and rephrase what you mean here, because I'm
> lost to understand ...... :)
I know that my English is VERY bad and I apologize. However, I will try
to explain again:
Now, Nimzo makes good moves, because CHE is telling what to do in
certain positions. Beeing a pre-processor (this word has become a kind
of invective by now) the program often shows a main line with bad moves.
So Chrilly tried to check CHE-Rules in every node of the search tree, no
matter what search depth. We expected the program to be slower and
smarter. But the program started to play bad moves, because it wanted to
get to a "good" position (good per CHE definition). To reach this "good"
position e.g. in the 9th ply it has to play bad moves. Another problem
was the cumulation of two or more CHE rules. Nimzo tried to reach
positions where it got bonuses from many CHE rules.
> OK, rephrase. Suppose you target a Q+B+N attack on h7, with
> the idea Qxh7 check and nasty follow ups.
> So you increase weightings for Q + B + N hitting h7, and the Queen
> gets exchanged. Still the B + N weights for the attack on h7 exist,
> but for no good reason anymore ...... ?
I agree that this kind of pre-processing can cause problems, but isn't a
chess program always a bunch of compromises? As I said before we are
working on CHE to solve this problems.
> But you must recognise that as the search space
> gets deeper the leaf node positions get further and further away
> from the ply 1 reality ?
Agreed. But in the early middle game it is important to move pieces to
certain squares, avoid certain pawn structures etc. etc. depending on
the opening. These squares and pawn structures are important for a
couple of moves, sometimes even for the endgame. When you have managed
to teach Nimzo how to play a opening well, you do not have to worry too
much about missing lines in the opening book!
>
> I was trying to make the point that this approach has a kind of
> build in redundancy (like car manufacture) such that the program
> does not maintain its improvement with larger search space as
> leaf-node processors ...... ?
>
We are working on this! I had a brilliant idea a few weeks ago... :)
I think, the point of KKs posting was that every user can implement
his/her own chess knowledge and this is a remarkable feature! I don't
know another commericial chess program where you can do this. Chrilly
and H.Weigel had to develop a whole language and a user interface. Sure,
CHE is not the answer to all chess problems, but it is amazing and it
will be improved!
Best wishes
Andreas Mader
Komputer Korner
mclane wrote:
>
snipped
> Some things die earlier than you will believe. Some things are even
> dead...
> >The komputer that kouldn't keep a password safe from
> >prying eyes, kouldn't kompute the square root of 36^n,
> >and kouldn't find the real motive in chessbase.
Before you pick a fight with me, please read carefully what I have said.
I did NOT say that Nimzo was the strongest program. A lot of you netters
seem to think that strength is everything. Well you are vastly
outnumbered
by most of us patzers that get the living daylights kicked out of us
by our computer programs. Strength is only one aspect of a program.
Look at CM5000. It will outsell all of the top programs. Extreme Chess
will also. Even Power Chess will. The real strength of Nimzo is
all the other aspects of the program including the world's first way
to apply chess knowledge through a menu interface. If you had even
tried out Nimzo for a hour, you would have fallen in love with it.
Strength is not everything even though Nimzo is not weak and
especially at fast time controls it is very good. It has a fantastic
opening book editor and it's database facilities are excellent as well.
Try it out. You will slowly come around to my opinion. So, it will never
be as strong as Rebel but it's rating on the SSDF is equivalent to
CM5000. When I said that it was superior in chess engine, what I meant
to
say was that it is superior in it's capability to change the chess
engine.
Your slanderous post was a result of a misunderstanding with what I
meant.
I didn't deserve your attack. I will forgive your strong words, if
you agree that we really are not in disagreement about strength of
Nimzo. It is not as strong as the top programs, but not so weak either.
Okay, peace declared or do you want to always make war?
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
mclane
Komputer Korner <kor...@netcom.ca> wrote:
>mclane wrote:
>>
>snipped
>> Some things die earlier than you will believe. Some things are even
>> dead...
>> >--
>> >The komputer that kouldn't keep a password safe from
>> >prying eyes, kouldn't kompute the square root of 36^n,
>> >and kouldn't find the real motive in chessbase.
>Before you pick a fight with me, please read carefully what I have said.
>I did NOT say that Nimzo was the strongest program. A lot of you netters
>seem to think that strength is everything. Well you are vastly
>outnumbered
>by most of us patzers that get the living daylights kicked out of us
>by our computer programs. Strength is only one aspect of a program.
>Look at CM5000. It will outsell all of the top programs. Extreme Chess
>will also. Even Power Chess will. The real strength of Nimzo is
>all the other aspects of the program including the world's first way
>to apply chess knowledge through a menu interface. If you had even
>tried out Nimzo for a hour, you would have fallen in love with it.
I have Nimzo. First I was very impressed but after I have seen games,
I was totally disapointed. Shall I post the games that disapoint me,
or do you not replay games at all?
>Strength is not everything even though Nimzo is not weak and
>especially at fast time controls it is very good. It has a fantastic
>opening book editor and it's database facilities are excellent as well.
>Try it out. You will slowly come around to my opinion.
It is a long time ago that I was surprised about Nimzo3.
You are quite late. Things take a long time to reach you, didn't they
?
> So, it will never
>be as strong as Rebel but it's rating on the SSDF is equivalent to
>CM5000. When I said that it was superior in chess engine, what I meant
>to
>say was that it is superior in it's capability to change the chess
>engine.
Here I disagree. The concept of fritz and nimzo is a dead-end-street.
If they try to put more knowledge into the search, there search
decreases to less positions, less plies and therefore they do not have
their main-strength anymore. A program that has strength through
search-depth reacts very sensible if you cut him search depth.
>Your slanderous post was a result of a misunderstanding with what I
>meant.
Sorry.
>I didn't deserve your attack. I will forgive your strong words, if
>you agree that we really are not in disagreement about strength of
>Nimzo.
We are.
I would prefer The King instead of Nimzo, when I analyse games or
retro-analyse games, BECAUSE THE KING EVALUATES IN THE TREE:
That is most important, from my experience.
A program that is a pre-processor can compute 60 hours about a
position, having 11 plies or even more search-depth, you input the
first move of the main-line and suddenly the whole evaluation changes!
I can't accept this for serious analysis.
Sorry. We cannot agree in this point. Although Johann has not made
much progress over the years (because he has not put much more
knowledge into his program - or?) I would prefer The king instead of
Fritz/Nimzo. You cannot trust their evaluation. It is based on search
and search is not chess.
It is proofed that logical-calculations lead to a paradoxon if you do
one after the other. In the end you get 1=0.
The aristoteles law from the excluded third element is not projecting
reality.
If I have a * (b + c) and this should be a*c + b*c this works in
maths but not in real life.
Also in maths Gödel said that this works not for well-ordered groups.
The whole mechanical - perspective is finished.
We are leaving the 20th century.
Fritz/Nimzo have lost this race. They are not winners. They were never
winners (despite of Hong Kong where it was 5 round and no opponents).
It's over. The next century will produce different needs and different
programs.
In 400 years they will consider about Fritz and Nimzo like we think
TODAY about calculators. We all have calculators. But they are nothing
that can be compared to a good symphony of Mahler or a good tasting
piece of chocolate. Calculators are machines. They are very precise.
But stupid. Mahler and e.g. chocolate is QUALITY.
The next century is not the century of QUANTITY like this century was.
Sorry that my words do not direct so much within the CHESS-WORLD, but
because this is very important for me, I ave to build bridges and give
examples.
>It is not as strong as the top programs, but not so weak either.
> Okay, peace declared or do you want to always make war?
>--
>Komputer Korner
Because I am a german ?!? :)
No - I refused the german-armee, so why should I fight with a
computer!!
Nono, we can live together and drink a beer without having the same
opinion. No problem. I don't need harmony.
>The komputer that kouldn't keep a password safe from
>prying eyes, kouldn't kompute the square root of 36^n,
>kouldn't find the real motive in ChessBase and missed
>the real learning feature of Nimzo. Long live Nimzo!!!!!!
Again: some things are even dead and nobody has found out...
mclane
"A.Mader" <andrea...@siemens.at> wrote:
>Chris Whittington wrote:
>> > I know that your program is no brute forcer and that it uses most of its
>> > knowledge in the whole search tree. Chrilly has tried to implement this,
>> > too. The result was that Nimzo played bad moves just because the good
>> > move was in the tree!
>>
>> Can you try and rephrase what you mean here, because I'm
>> lost to understand ...... :)
>I know that my English is VERY bad and I apologize. However, I will try
>to explain again:
>Now, Nimzo makes good moves, because CHE is telling what to do in
>certain positions. Beeing a pre-processor (this word has become a kind
>of invective by now) the program often shows a main line with bad moves.
>So Chrilly tried to check CHE-Rules in every node of the search tree, no
>matter what search depth. We expected the program to be slower and
>smarter. But the program started to play bad moves, because it wanted to
>get to a "good" position (good per CHE definition). To reach this "good"
>position e.g. in the 9th ply it has to play bad moves. Another problem
>was the cumulation of two or more CHE rules. Nimzo tried to reach
>positions where it got bonuses from many CHE rules.
Looks like NIMZO is not a good strategist if it plays bad moves to
reach good ones !!
I read in ComputerSchach & Spiele it is a strategist !!
Hm - it has to play bad moves to reach the good position ???
Do you understand Chris ?
It looks these austrians do TOTALLY other things than we have done
over the years ?!?
Now I know why they need to play 5000 or more games ...
They have a strange method of WORKING on the program....
Now I understand what a NULL-MOVE is :-)
>> OK, rephrase. Suppose you target a Q+B+N attack on h7, with
>> the idea Qxh7 check and nasty follow ups.
>> So you increase weightings for Q + B + N hitting h7, and the Queen
>> gets exchanged. Still the B + N weights for the attack on h7 exist,
>> but for no good reason anymore ...... ?
>I agree that this kind of pre-processing can cause problems, but isn't a
>chess program always a bunch of compromises? As I said before we are
>working on CHE to solve this problems.
Solve it !! I will update then!!
BTW: Jakarta version of Nimzo plays not much better chess.
Maybe it FINDS earlier in some tactical positions, but against an
intelligent program it plays really unsound.
>
>> But you must recognise that as the search space
>> gets deeper the leaf node positions get further and further away
>> from the ply 1 reality ?
>Agreed. But in the early middle game it is important to move pieces to
>certain squares, avoid certain pawn structures etc. etc. depending on
>the opening. These squares and pawn structures are important for a
>couple of moves, sometimes even for the endgame. When you have managed
>to teach Nimzo how to play a opening well, you do not have to worry too
>much about missing lines in the opening book!
You teach NIMZO to play the pieces onto certain squares, and after
that it rebuilds everything because it has no further criteria to go
on itself.
Thats SYSIPHUS work.
You can't write a chess program alone with IF-clauses.
>>
>> I was trying to make the point that this approach has a kind of
>> build in redundancy (like car manufacture) such that the program
>> does not maintain its improvement with larger search space as
>> leaf-node processors ...... ?
>>
>We are working on this! I had a brilliant idea a few weeks ago... :)
Brilliant! Me too!
>I think, the point of KKs posting was that every user can implement
>his/her own chess knowledge and this is a remarkable feature! I don't
>know another commericial chess program where you can do this. Chrilly
>and H.Weigel had to develop a whole language and a user interface. Sure,
>CHE is not the answer to all chess problems, but it is amazing and it
>will be improved!
Indeed this idea was remarkable. If we could program this knowledge
for the whole tree without the effect that NIMZO PLAYS WEAK MOVES
TO REACH THE GOOD POSITION (is this alpha-beta :-) )))
we all could work on evaluations.
As long as the knowledge it MISUSED, we have to wait until ...
next century ???
>Best wishes
>Andreas Mader
mclane
"A.Mader" <andrea...@siemens.at> wrote:
>I know that my English is VERY bad and I apologize. However, I will try
>to explain again:
>Now, Nimzo makes good moves, because CHE is telling what to do in
>certain positions. Beeing a pre-processor (this word has become a kind
>of invective by now) the program often shows a main line with bad moves.
>So Chrilly tried to check CHE-Rules in every node of the search tree, no
>matter what search depth. We expected the program to be slower and
>smarter. But the program started to play bad moves, because it wanted to
>get to a "good" position (good per CHE definition). To reach this "good"
>position e.g. in the 9th ply it has to play bad moves. Another problem
>was the cumulation of two or more CHE rules. Nimzo tried to reach
>positions where it got bonuses from many CHE rules.
Aha - now I understand. Why don't you explain the reason instead of
decribing the phenomena:
Child=Nimzo learns that after it cried it gets one MARS nor snickers
to heal it's pain.
Next step child does is, it cries all the way to get more and more
MARS or snickers.
But there is an easy way to handle this .... we had the same problem
in CSTal.
Opponent got a double pawn. Instead of eating this pawn CSTAL tiried
to collect more and more maluses of the opponent.
A kind of knowledge-horizont-problem.
If you consider about the child in my anology, you can find an easy
way to handle it....
Sometimes life gives us ideas, ... ideas enough.
Ed Schröder
I don't think it's a good idea to fight a personal vendetta with as
victim a chess program.
Thorsten?
Please...
- Ed -
mcl...@prima.ruhr.de (mclane) wrote:
>"A.Mader" <andrea...@siemens.at> wrote:
>
>>Chris Whittington wrote:
>>> > I know that your program is no brute forcer and that it uses most of its
>>> > knowledge in the whole search tree. Chrilly has tried to implement this,
>>> > too. The result was that Nimzo played bad moves just because the good
>>> > move was in the tree!
>>>
>>> Can you try and rephrase what you mean here, because I'm
>>> lost to understand ...... :)
>
>>I know that my English is VERY bad and I apologize. However, I will try
>>to explain again:
>>Now, Nimzo makes good moves, because CHE is telling what to do in
>>certain positions. Beeing a pre-processor (this word has become a kind
>>of invective by now) the program often shows a main line with bad moves.
>>So Chrilly tried to check CHE-Rules in every node of the search tree, no
>>matter what search depth. We expected the program to be slower and
>>smarter. But the program started to play bad moves, because it wanted to
>>get to a "good" position (good per CHE definition). To reach this "good"
>>position e.g. in the 9th ply it has to play bad moves. Another problem
>>was the cumulation of two or more CHE rules. Nimzo tried to reach
>>positions where it got bonuses from many CHE rules.
>
>
>Looks like NIMZO is not a good strategist if it plays bad moves to
>reach good ones !!
>
>I read in ComputerSchach & Spiele it is a strategist !!
>
>Hm - it has to play bad moves to reach the good position ???
>
>Do you understand Chris ?
>
>It looks these austrians do TOTALLY other things than we have done
>over the years ?!?
>Now I know why they need to play 5000 or more games ...
>They have a strange method of WORKING on the program....
>Now I understand what a NULL-MOVE is :-)
>
>>> OK, rephrase. Suppose you target a Q+B+N attack on h7, with
>>> the idea Qxh7 check and nasty follow ups.
>>> So you increase weightings for Q + B + N hitting h7, and the Queen
>>> gets exchanged. Still the B + N weights for the attack on h7 exist,
>>> but for no good reason anymore ...... ?
>
>>I agree that this kind of pre-processing can cause problems, but isn't a
>>chess program always a bunch of compromises? As I said before we are
>>working on CHE to solve this problems.
>
>
>Solve it !! I will update then!!
>BTW: Jakarta version of Nimzo plays not much better chess.
>Maybe it FINDS earlier in some tactical positions, but against an
>intelligent program it plays really unsound.
>
>>
>>> But you must recognise that as the search space
>>> gets deeper the leaf node positions get further and further away
>>> from the ply 1 reality ?
>
>>Agreed. But in the early middle game it is important to move pieces to
>>certain squares, avoid certain pawn structures etc. etc. depending on
>>the opening. These squares and pawn structures are important for a
>>couple of moves, sometimes even for the endgame. When you have managed
>>to teach Nimzo how to play a opening well, you do not have to worry too
>>much about missing lines in the opening book!
>
>
>You teach NIMZO to play the pieces onto certain squares, and after
>that it rebuilds everything because it has no further criteria to go
>on itself.
>Thats SYSIPHUS work.
>You can't write a chess program alone with IF-clauses.
>
>>>
>>> I was trying to make the point that this approach has a kind of
>>> build in redundancy (like car manufacture) such that the program
>>> does not maintain its improvement with larger search space as
>>> leaf-node processors ...... ?
>>>
>
Tord Kallqvist Romstad
mclane (mcl...@prima.ruhr.de) wrote:
<snip>
: Here I disagree. The concept of fritz and nimzo is a dead-end-street.
: If they try to put more knowledge into the search, there search
: decreases to less positions, less plies and therefore they do not have
: their main-strength anymore. A program that has strength through
: search-depth reacts very sensible if you cut him search depth.
: >Your slanderous post was a result of a misunderstanding with what I
: >meant.
: Sorry.
: >I didn't deserve your attack. I will forgive your strong words, if
: >you agree that we really are not in disagreement about strength of
: >Nimzo.
: We are.
: I would prefer The King instead of Nimzo, when I analyse games or
: retro-analyse games, BECAUSE THE KING EVALUATES IN THE TREE:
: That is most important, from my experience.
: A program that is a pre-processor can compute 60 hours about a
: position, having 11 plies or even more search-depth, you input the
: first move of the main-line and suddenly the whole evaluation changes!
: I can't accept this for serious analysis.
: Sorry. We cannot agree in this point. Although Johann has not made
: much progress over the years (because he has not put much more
The King 2.5 is _much_ stronger than the King 1.0.
: knowledge into his program - or?) I would prefer The king instead of
: Fritz/Nimzo. You cannot trust their evaluation. It is based on search
: and search is not chess.
Of course not. Search is not chess. An evaluation function, no matter how
complicated, is not chess. Chess is a game, not the algorithms or thought
processes we use to play it. And you are right, you can't trust the
evaluation of Fritz/Nimzo, but you can also not trust the evaluation of
any other chess programs. There is even less reason to trust the evaluation
of your beloved CST, according to Chris. The only thing you can rely on a
chess program to find is tactics, and Fritz / Nimzo will usually find
these faster than most other programs. For serious analysis Genius would
be my first choice, Fritz the second. I don't want my programs to find
strong positional moves, because I do that better myself anyway. When
playing games, of course, everything is different. The fast searchers are
usually less fun to play against.
: It is proofed that logical-calculations lead to a paradoxon if you do
: one after the other. In the end you get 1=0.
: The aristoteles law from the excluded third element is not projecting
: reality.
: If I have a * (b + c) and this should be a*c + b*c this works in
: maths but not in real life.
: Also in maths Gödel said that this works not for well-ordered groups.
What are you talking about? I am a graduate maths student, and I don't
understand a single word of your last paragraph. I also fail to see what
this has to do with Fritz/Nimzo.
: The whole mechanical - perspective is finished.
: We are leaving the 20th century.
: Fritz/Nimzo have lost this race. They are not winners. They were never
: winners (despite of Hong Kong where it was 5 round and no opponents).
5 rounds is of course not enough, but no opponents? I consider Deep Thought
to be quite a strong opponent...
: It's over. The next century will produce different needs and different
: programs.
: In 400 years they will consider about Fritz and Nimzo like we think
: TODAY about calculators.
In 400 years we will all be dead, and Fritz, Nimzo, the King, CST,
Bob Hyatt, Richard Lang and Ed Schroder will all be forgotten.
: We all have calculators. But they are nothing
: that can be compared to a good symphony of Mahler or a good tasting
: piece of chocolate. Calculators are machines. They are very precise.
: But stupid. Mahler and e.g. chocolate is QUALITY.
Mahler didn't write any good symphonies. :-)
I prefer chocolate (or Bach).
Here is a part of my "quality spectrum": :-)
Much quality <-----------------------------------------------> Little quality
Bach Gödel Rebel The King Coffee Fritz Chocolate Mahler Thorsten's
postings
in rgcc
Seriously, comparing the quality of composers, calculators, chess programs
and chocolate doesn't seem very fruitful to me.
Also, some of your postings are actually very interesting!
: The next century is not the century of QUANTITY like this century was.
: Sorry that my words do not direct so much within the CHESS-WORLD, but
: because this is very important for me, I ave to build bridges and give
: examples.
It's OK that you words "do not direct so much within the CHESS-WORLD".
The problem is that they don't seem to make any sense at all.
: >It is not as strong as the top programs, but not so weak either.
: > Okay, peace declared or do you want to always make war?
: >--
: >Komputer Korner
: Because I am a german ?!? :)
: No - I refused the german-armee, so why should I fight with a
^^^^^^^^^^^^^^^^^^^^^^^^^^
Wise decision! Nice that we can agree about something!
: computer!!
: Nono, we can live together and drink a beer without having the same
: opinion.
Sounds nice, I hope I can join you (perhaps my program will be ready
for play in AEGON next year...). :-)
Tord
: No problem. I don't need harmony.
Chris Whittington
Thorsten, I'm going to flame you for this.
It's not better than the attacks on the Americans.
I explain for those who don't understand.
Some Germans refer to foreigners as apes (the word untermensch
being no longer politically acceptable).
Those that know, know that (some) Germans refer to Austrians
as apen (apes). The English are inseln-apen, island-apes :)
The earliest reference to this comes from Bismarck.
Asked his opinion about Bavarians, he replied: "sie sind halfweg
zwischen Ostereicher und mensch" - which translates to "they are
halway between the Austrians and humans", namely half-apes - Bismarck
himself was a Prussian (mensch, human), and correct in the
geographical sense.
So why is the communist, internationalist, Thorsten resorting
to right-wing Prussian nationalism, using 'apes' ? Especially when
Thorsten is not really German at all (Czub is no German name,
its Bohemian, Czeckoslovak).
To try and 'subtly' wind up the Austrian, A.Mader ?
To try and become a 'true' German ?
Come on, Thorsten, stop this racist, nationalist stuff,
and start some sensible criticism that we can engage in a
sensible dialog over.
Chris Whittington
Ed Schröder
Fact remains that Nimzo3 made a remarkable ELO jump compared to Nimzo2.
- Ed -
mclane
Chris Whittington <chr...@cpsoft.demon.co.uk> wrote:
>> Randomnly played chess!
>> Chess by drunken apes!!
>Thorsten, I'm going to flame you for this.
>It's not better than the attacks on the Americans.
>I explain for those who don't understand.
For me....
>Some Germans refer to foreigners as apes (the word untermensch
>being no longer politically acceptable).
AM I such a german in your eyes ?? Be precise!
>Those that know, know that (some) Germans refer to Austrians
>as apen (apes). The English are inseln-apen, island-apes :)
Oh - now I see the fault. I have told chris once that english-guys are
known as island-apes in gemany. But I did not told him that
I THINK THIS. Because I like britain and would like to live there.
But he has not said that I SAID ISLAND_APES. OK. I CALM DOWN...
>The earliest reference to this comes from Bismarck.
>Asked his opinion about Bavarians, he replied: "sie sind halfweg
>zwischen Ostereicher und mensch" - which translates to "they are
>halway between the Austrians and humans", namely half-apes - Bismarck
>himself was a Prussian (mensch, human), and correct in the
>geographical sense.
Bismarck was a motherson. I don't like him.
You know more about Bismarck than me, chris.
Only racists or fashists would quote this stuff about bismarck.
I don't know you are not a racist and also no fashist. Only a
capitlaist.
So : what shall this flame tell me ?
That you like to make jokes about the prejudices of SOME germans,
because you lived in germany and liked the culture ?!
(and the intelligent german girls)
>So why is the communist, internationalist, Thorsten resorting
>to right-wing Prussian nationalism, using 'apes' ? Especially when
>Thorsten is not really German at all (Czub is no German name,
>its Bohemian, Czeckoslovak).
13 generations. And I was asked when I refused german armee if CZUB is
a arish name ??!?
I am a racist because i call somebody an ape ?????
This is a joke , isn't it ???
AGAIN: I do not speak in terms like:
Österreicher sind Affen, Engländer sind INselaffen, Italiener sind
Itakker etc etc.
because I have nothing against austrians, britains and and and.
I have prejudice against the american CULTURE .
Don't you go to much forward?
So anybody using the word ape in the context of :
You are an ape! is a right-wing- prussian ????
Hm . Thats sound a little drunken to me.
>To try and 'subtly' wind up the Austrian, A.Mader ?
>To try and become a 'true' German ?
>Come on, Thorsten, stop this racist, nationalist stuff,
>and start some sensible criticism that we can engage in a
>sensible dialog over.
NONONO. You started the ape discussion as a racistic discussion.
I forgive you because I don't understand this FLAME at all.
It is not my style to say these racistic statements you say SOME
germans say. With whom have you spoken ????
I know that there are many germans that have prejudices against
foreigners and that they are nationalists and fashists.
I know these germans. Some are even in my family.
E.g. my father was always the one calling italians ITAKKER and
britains island-apes. Of course my father is right-wing. And he is
electing CDU. But I was never much impressed by my father.
When I attack Andreas, or Nimzo, than of course not because he is
austrian. He could be indonesian and I would attack him.
I have some prejudices against american culture.
And many about germans. I hate germans.
Especially west-germans.
mclane
"Ed Schröder" <rebc...@xs4all.nl> wrote:
>I don't think it's a good idea to fight a personal vendetta with as
>victim a chess program.
>Thorsten?
>Please...
>- Ed -
To make this clear:
I have nothing against Chrilly or Nimzo.
I am just saying that it is not playing chess.
I have also nothing against Andreas Mader.
You don't need to defend them.
I have not attacked them, but what they do.
Thats different.
I comment on their work, not on the persons.
I like Chrilly. He is very sympathic. The first time I met him in
paderborn, a few years ago. And from the first time I liked him, he
was so engaged in the game, why should I attack HIM ?
You defend because he is now a professional.
In the ROCHADE -EUROPA Karsten Bauermeister wrote:
"Nimzo 3 mit Bediener Helmut Weigel meldete noch als Amateur , weil
Nimzo zum Zeitpunkt der Anmeldung noch nicht kommerziell erhältlich
war. "
In english:
Nimzo was reported as an amateur because Nimzo was not commercially
available before the championship .
When I read such a lie, or misunderstanding, or whatever, I could
attack this.
But I do not attack Chrilly.
Fact is: Nimzo WAS commercially available. Many programs in the past
before were not commercially available BEFORE the championship.
Thats normal. If they all are AMATEURS, than Richard Lang is an
amateur too.
Mephisto Amsterdam was not available before the championship.
Mephisto Dallas ....
Mephisto Roma ....
and and and.
So :
Why don't say the truth:
They had not enough money or don't wanted to spent so much money for
flying to Jakarta.
So they though: we could try to cheat saying: we are not commercially
available, therefore we are amateurs.
And it worked.
But the above sentence is stupis.
I don't know if Karsten Bauermeister made the mistake,
the ICCA, the Nimzo-team or whoever.
But when I comment on the sentence, I comment on what they have done,
and not flaming the persons or attacking the persons.
Anyone can understand that a new company has not that much money, of
course if the program is not that strong like the first-reviews wanted
to suggest.
BUT: I don't accept lies.
You don't make the things more wizard when you fool the people here
with the sentence that I have a personal-fight with somebody and am
therefore not serious.
Thats not true.
You don't need to defend NIMZO.
Maybe some people should try to work fair and to publish without
lying.
Then maybe they would have more success !!
If CSS said NIMZO is a SUPER program (KK repeated this nonsense a few
weeks ago. I could speak in terms of Andreas and say: is KK maybe an
alias for Andreas Mader, or say: Greeting to AM dear KK. But I am not
paranois like him) they have not said that much the truth.
When Nimzo is now not selling the best, they should try it next time
with the truth !
Maybe the people are not that stupid like Steinwender thinks they are.
If Chrilly flies to Jakarta because his advisers have told him:
we have a chance to make a title
and he fails , his advisors are wrong!
Thats not my problem. More an inner-austrian-desease.
Maybe they are like chessbase:
not having the competence, not having the resources.
(I am just quoting what wuellenweber said!!!).
Again my suggestion: change methods!
Chris Whittington
mcl...@prima.ruhr.de (mclane) wrote:
>
> Chris Whittington <chr...@cpsoft.demon.co.uk> wrote:
>
>
> >> Randomnly played chess!
> >> Chess by drunken apes!!
>
> >Thorsten, I'm going to flame you for this.
>
> >It's not better than the attacks on the Americans.
>
> >I explain for those who don't understand.
>
> For me....
>
> >Some Germans refer to foreigners as apes (the word untermensch
> >being no longer politically acceptable).
>
> AM I such a german in your eyes ?? Be precise!
Precise ? Ok, the answer is 'NO'. :)
>
> >Those that know, know that (some) Germans refer to Austrians
> >as apen (apes). The English are inseln-apen, island-apes :)
>
> Oh - now I see the fault. I have told chris once that english-guys are
> known as island-apes in gemany. But I did not told him that
> I THINK THIS. Because I like britain and would like to live there.
> But he has not said that I SAID ISLAND_APES. OK. I CALM DOWN...
>
> >The earliest reference to this comes from Bismarck.
> >Asked his opinion about Bavarians, he replied: "sie sind halfweg
> >zwischen Ostereicher und mensch" - which translates to "they are
> >halway between the Austrians and humans", namely half-apes - Bismarck
> >himself was a Prussian (mensch, human), and correct in the
> >geographical sense.
>
> Bismarck was a motherson. I don't like him.
> You know more about Bismarck than me, chris.
Yup, here in England, we make a speciality of German history.
Germany is our deadliest enemy, so we need to understand Germans :)
> Only racists or fashists would quote this stuff about bismarck.
This is silly putting people into boxes stuff.
You know everybody is at the same time a racist and a faschist
and a communist and a consumer and a worker and and and.
> I don't know you are not a racist and also no fashist. Only a
> capitlaist.
Yup, and I believe in profit :)
> So : what shall this flame tell me ?
> That you like to make jokes about the prejudices of SOME germans,
> because you lived in germany and liked the culture ?!
> (and the intelligent german girls)
Careful, maybe my wife reads this :)
>
>
> >So why is the communist, internationalist, Thorsten resorting
> >to right-wing Prussian nationalism, using 'apes' ? Especially when
> >Thorsten is not really German at all (Czub is no German name,
> >its Bohemian, Czeckoslovak).
>
> 13 generations. And I was asked when I refused german armee if CZUB is
> a arish name ??!?
>
> I am a racist because i call somebody an ape ?????
> This is a joke , isn't it ???
> AGAIN: I do not speak in terms like:
> Österreicher sind Affen, Engländer sind INselaffen, Italiener sind
> Itakker etc etc.
> because I have nothing against austrians, britains and and and.
> I have prejudice against the american CULTURE .
> Don't you go to much forward?
> So anybody using the word ape in the context of :
> You are an ape! is a right-wing- prussian ????
>
> Hm . Thats sound a little drunken to me.
Just another 'crazy' idea :)
> >To try and 'subtly' wind up the Austrian, A.Mader ?
>
> >To try and become a 'true' German ?
>
> >Come on, Thorsten, stop this racist, nationalist stuff,
> >and start some sensible criticism that we can engage in a
> >sensible dialog over.
>
> NONONO. You started the ape discussion as a racistic discussion.
> I forgive you because I don't understand this FLAME at all.
> It is not my style to say these racistic statements you say SOME
> germans say. With whom have you spoken ????
> I know that there are many germans that have prejudices against
> foreigners and that they are nationalists and fashists.
> I know these germans. Some are even in my family.
> E.g. my father was always the one calling italians ITAKKER and
> britains island-apes. Of course my father is right-wing. And he is
> electing CDU. But I was never much impressed by my father.
>
> When I attack Andreas, or Nimzo, than of course not because he is
> austrian. He could be indonesian and I would attack him.
>
> I have some prejudices against american culture.
This is ok. So do I. So do most european intellectuals.
But not on r.g.c.c.
> And many about germans. I hate germans.
This is not ok. german *culture*, ok. But germans, not ok.
But you are unfortunately not alone.
> Especially west-germans.
Try adding culture to the above.
Otherwise I'll get to thinking that you believe in the german gene,
passed from generation to generation - and causing all sorts
of problems to the neighbours. Now that would be racist, no ?
But; ok, it was an over-the-top reaction from me.
I apologise.
Maybe what I'm trying to say is that we should all stop the
references to the national origin of posters.
It doesn't matter who is german, or US, or austrian or english.
All that happens when this language is used is that someone (or
many) get annoyed; and then the original r.g.c.c. meaning is
lost in the storm.
If in Rome, then one does as the Romans. Speak Italian.
In r.g.c.c., then speak computer chess language.
You can make evolutionary changes to the language and make
different the topics that are 'on-topic'.
But revolutionary additions, whether by abuse, or nationalism,
or whatever, just don't work.
And, Thorsten, I know you are very clever and perceptive about
computer chess (and other issues). Its a shame that a lot of
what you post can be ignored or gets lost because of this
revolutionary language.
Most of these r.g.c.c. guys are quite conservative. Many of them maybe
just did engineering or computer science. They didn't do
loads of philosophy and politics and history, and, certainly, the
American ones for sure, don't go through the same sort of
Marxist mist that we get here in europe by being a student/refusenik
or whatever. So if you talk of Karl Liebnitz, they think he
manufactures watches, or Rosa Luxembourg is a pop star; they don't
operate with the same set of ideas and factoids and words
that you have. And if you use these ideas in posts they'll just
reckon you are nuts.
I'ld like to see your on-topic posts answered (more often), because
you make very useful observations.
I'ld like to see the off-topic flame wars subside and go away. Not
because I want them censored, but because too much off-topic
stuff from one person results in that person being generally
ignored, even when they talk sense.
Your friend,
Chris Whittington
Enrique Irazoqui
Chris Whittington <chr...@cpsoft.demon.co.uk> wrote in article
<85050787...@cpsoft.demon.co.uk>...
> You know everybody is at the same time a racist and a faschist
> and a communist and a consumer and a worker and and and.
No way! Certainly not fascist and not racist, Enrique
mclane
Chris Whittington <chr...@cpsoft.demon.co.uk> wrote:
>mcl...@prima.ruhr.de (mclane) wrote:
>>
>> Chris Whittington <chr...@cpsoft.demon.co.uk> wrote:
>>
>>
>> >Some Germans refer to foreigners as apes (the word untermensch
>> >being no longer politically acceptable).
>>
>> AM I such a german in your eyes ?? Be precise!
>Precise ? Ok, the answer is 'NO'. :)
Thanks (at all) !
>>
>> >Those that know, know that (some) Germans refer to Austrians
>> >as apen (apes). The English are inseln-apen, island-apes :)
>>
>> Bismarck was a motherson. I don't like him.
>> You know more about Bismarck than me, chris.
>Yup, here in England, we make a speciality of German history.
>Germany is our deadliest enemy, so we need to understand Germans :)
How can we be friends if germany is in war (again) with britain ?
>> Only racists or fashists would quote this stuff about bismarck.
>This is silly putting people into boxes stuff.
>You know everybody is at the same time a racist and a faschist
>and a communist and a consumer and a worker and and and.
No - I have different opinion. sorry.
>> I don't know you are not a racist and also no fashist. Only a
>> capitlaist.
>Yup, and I believe in profit :)
You a capitalist, me a communist. You an english guy, me a german. How
can this fit together ?!
It fits. Voila ! C'est l'europe !
The united states of europe (but whom shall we sell the
mad-cow-desease-meat, maybe we should give it to the americans.
Mad-Cow-Hamburgers!
>> So : what shall this flame tell me ?
>> That you like to make jokes about the prejudices of SOME germans,
>> because you lived in germany and liked the culture ?!
>> (and the intelligent german girls)
>Careful, maybe my wife reads this :)
BTW: When do I met your nice wife ?! She does not seem to have an easy
job! 6 girls and one Chris Whittington? A kind of british copy of the
Kelly-Family. Chris with grey hair and always short before a stroke.
Horrible to consider about.
>>
>> I am a racist because i call somebody an ape ?????
[...]
>> Don't you go to much forward?
>> So anybody using the word ape in the context of :
>> You are an ape! is a right-wing- prussian ????
>>
>> Hm . Thats sound a little drunken to me.
>Just another 'crazy' idea :)
>> >To try and 'subtly' wind up the Austrian, A.Mader ?
>>
>> >To try and become a 'true' German ?
>>
>> >Come on, Thorsten, stop this racist, nationalist stuff,
>> >and start some sensible criticism that we can engage in a
>> >sensible dialog over.
>>
[...]
>>
>> When I attack Andreas, or Nimzo, than of course not because he is
>> austrian. He could be indonesian and I would attack him.
>>
>> I have some prejudices against american culture.
>This is ok. So do I. So do most european intellectuals.
>But not on r.g.c.c.
Prejudices are not very good to control. But better speak them out
instead of thinking and having a double-moral.
>> And many about germans. I hate germans.
>This is not ok. german *culture*, ok. But germans, not ok.
>But you are unfortunately not alone.
Yes. And that makes me feel very good.
>> Especially west-germans.
>Try adding culture to the above.
>Otherwise I'll get to thinking that you believe in the german gene,
>passed from generation to generation - and causing all sorts
>of problems to the neighbours. Now that would be racist, no ?
I don't believe in genetical stuff. Materialist do.
>But; ok, it was an over-the-top reaction from me.
>I apologise.
Thanks.
>Maybe what I'm trying to say is that we should all stop the
>references to the national origin of posters.
>It doesn't matter who is german, or US, or austrian or english.
Right. This is the reason I am mclane, he was the commander of the
starship ORION, and lived in the 30.century.
>All that happens when this language is used is that someone (or
>many) get annoyed; and then the original r.g.c.c. meaning is
>lost in the storm.
>If in Rome, then one does as the Romans. Speak Italian.
>In r.g.c.c., then speak computer chess language.
I would like to speak in games, and show results and notations.
But experience has showed: nobody replays chess-games anymore.
They are all too lazy , some have even forgotten about notation.
Instead of this they have multi-media sound and mpeg-films of
stripping girls and that stuff.
>You can make evolutionary changes to the language and make
>different the topics that are 'on-topic'.
>But revolutionary additions, whether by abuse, or nationalism,
>or whatever, just don't work.
Capitalism will fail, just 3 more years and you will see!
>And, Thorsten, I know you are very clever and perceptive about
>computer chess (and other issues). Its a shame that a lot of
>what you post can be ignored or gets lost because of this
>revolutionary language.
Maybe. But having double-moral is not better (now I don't speak about
you, Chris.). I have met so many NICE and polite guys, I am rid of
these parasites. They live from things/ideas others invent, they smile
and are diplomatical, and lie the rest of the day. They exploit and
tell their slaves: come on - we are all one team, one family.
All that to get more power, more exploitation. No - I want to live
honest and fair with people.
"We are spirits, in a material world..."
>Most of these r.g.c.c. guys are quite conservative. Many of them maybe
>just did engineering or computer science. They didn't do
>loads of philosophy and politics and history, and, certainly, the
>American ones for sure, don't go through the same sort of
>Marxist mist that we get here in europe by being a student/refusenik
>or whatever. So if you talk of Karl Liebnitz, they think he
>manufactures watches, or Rosa Luxembourg is a pop star; they don't
>operate with the same set of ideas and factoids and words
>that you have. And if you use these ideas in posts they'll just
>reckon you are nuts.
Ok - you are right.
>I'ld like to see your on-topic posts answered (more often), because
>you make very useful observations.
>I'ld like to see the off-topic flame wars subside and go away. Not
>because I want them censored, but because too much off-topic
>stuff from one person results in that person being generally
>ignored, even when they talk sense.
>Your friend,
>Chris Whittington
All I can answer to this is, that I like Mchess for many years now,
I've met the programmer once, and I was upset and angry when I saw the
anti-mchess-campaign.
The same thing happened later when somebody attacked Ed.
It happens also when somebody says strange things about CSTAL or about
you.
I can't help. I once worked in a chess-computer magazine, it was the
same problem. I felt too much for the programmers, was to much hurted
when I saw the intriques and the dirty-tricks-behind the scene.
When I like a programmer, and I have worked months or more with him,
maybe a kind of relationship has build, it hurts to see that
COMMERCIAL-GUYS, not interested in computer-chess, but in making
business, betray these programmers with contracts, breaking-contracts
or bad-press, attacking their programs, writing mean articles about
their programs AND AND AND.
That hurts me.
It was never my intention to hurt Goran. I am sorry that he has
written his letter of parting.
Also what I said about americans was not said to attack bruce, or bob.
Why don't you want to understand:
here in germany we have a war. This is a big computer-chess-market.
And they work with dirty-tricks.
When I meet the programmers somewhere, it is always a nice atmoshere,
we alltogether join computerchess and have a nice day.
But the things change when money-interests poison the event. Suddenly
you are in the bad-company of those business-guys.
And suddenly the "commercial-programmers" have to fight against others
just to save their life. To earn enough. Suddenly moral becomes
flexible and suddenly gossip is spread.
I know this has not much to do with computerchess. Sure.
But on the other hand, it has something to do with computerchess.
Maybe the amateurs will understand what I said when they are
commercials.
I apologize for these OFF-TOPICs.
Komputer Korner
mclane wrote:
>
snipped
> When I meet the programmers somewhere, it is always a nice atmoshere,
> we alltogether join computerchess and have a nice day.
> But the things change when money-interests poison the event. Suddenly
> you are in the bad-company of those business-guys.
> And suddenly the "commercial-programmers" have to fight against others
> just to save their life. To earn enough. Suddenly moral becomes
> flexible and suddenly gossip is spread.
>
> I know this has not much to do with computerchess. Sure.
> But on the other hand, it has something to do with computerchess.
> Maybe the amateurs will understand what I said when they are
> commercials.
>
> I apologize for these OFF-TOPICs.
Thorsten, I fell in love with Nimzo because of the features, not because
it calculates at the root. Maybe the more I play with it, I will come
to realize that it's root evaluation is just too much of a handicap.
but for now I am like a kid in a candy store because of its' features.
I am not in anybody's pocket, just a komputer addicted to chess.
Try not to be so harsh on me!!! I will publish the definitive review of
Nimzo including the root evaluation controversy. Who knows maybe Nimzo
is
a mirage, but for now what a mirage!!!
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
mclane
Komputer Korner <kor...@netcom.ca> wrote:
>mclane wrote:
>>
>snipped
>> I apologize for these OFF-TOPICs.
>Thorsten, I fell in love with Nimzo because of the features, not because
>it calculates at the root. Maybe the more I play with it, I will come
>to realize that it's root evaluation is just too much of a handicap.
>but for now I am like a kid in a candy store because of its' features.
>I am not in anybody's pocket, just a komputer addicted to chess.
>Try not to be so harsh on me!!! I will publish the definitive review of
>Nimzo including the root evaluation controversy. Who knows maybe Nimzo
>is
>a mirage, but for now what a mirage!!!
>--
>Komputer Korner
Nimzo has many nice features and a great database function. There are
many many nice details in the program. It is programmed with much love
for deatils, you can see that a team of advisors like Thomas, Helmut,
Andreas and all the others have helped Chrilly to create such a
professional tool for chess!
There is no doubt about this. If they would have published the
program, it would have given them much money and reputation.
But than they made the deal with Mr. Steinwender, and they published
nasty things in the magazine. And that is unfair.
If your program IS NOT THAT STRONG, and you publish everywhere:
Oh - it is amazingly strong.
Then you have a problem.
That is the point I refer to.
If they would have published it without this advertising machinery,
everything would have been ok.
I liked Nimzo the same like you liked it. But within a few days I
found out that nimzo is A LITTLE STRONGER THAN FRITZ3, not less and
not more. And then I though: why do they do as if Nimzo is much
stronger when it isn't.
>The komputer that kouldn't keep a password safe from
>prying eyes, kouldn't kompute the square root of 36^n,
Harald Faber
Hello mclane,
I think I have an interesting aspect to tell you:
m> I have Nimzo. First I was very impressed but after I have seen games,
m> I was totally disapointed. Shall I post the games that disapoint me,
m> or do you not replay games at all?
Go for it, proves are always better than claims.
Ciao and see ya
Harald
--
mclane
Harald...@p21.f2.n1.z1001.fidonet.org (Harald Faber) wrote:
>Hello mclane,
Ok, I will show you the weak games of nimzo3 and nimzo3.5 .
For the meanthime you can study the match between Hiarcs5 and Nimzo
Jakarta 3.5 in another thread.
Harald Faber
Hello mclane,
m> Ok, I will show you the weak games of nimzo3 and nimzo3.5 .
m> For the meanthime you can study the match between Hiarcs5 and Nimzo
m> Jakarta 3.5 in another thread.
Hmm, I'd like to see results from MCP6 against Hiarcs5 and Nimzo3.5,
especially with tournament-conditions, not only 60moves/60mins.
I'd also like to see MCP6 play against Wchess which seems to be very
strong, at least in our German Chessnet when playing via e-mail, and
against The King 2.54. Can anybody effort to get results maybe from
autoplayer?
mclane
Harald...@p21.f2.n1.z1001.fidonet.org (Harald Faber) wrote:
>Hello mclane,
>m> Ok, I will show you the weak games of nimzo3 and nimzo3.5 .
>m> For the meanthime you can study the match between Hiarcs5 and Nimzo
>m> Jakarta 3.5 in another thread.
>Hmm, I'd like to see results from MCP6 against Hiarcs5 and Nimzo3.5,
We are working on this....
>especially with tournament-conditions, not only 60moves/60mins.
we are even more working on that. But - as you know - it takes some
hours even for my autoplayers.
>I'd also like to see MCP6 play against Wchess which seems to be very
>strong, at least in our German Chessnet when playing via e-mail, and
>against The King 2.54. Can anybody effort to get results maybe from
>autoplayer?
I would like to, but my last token of Wchess is on my 486-DX4 100 Mhz
PC. Because the token is not uninstallable, and I don't want to buy
the program twice, .... shit copy protection. I have nothing against
copy protection. But if the tokens are not uninstallable and when I
have no support from a dealer...
Harald Faber
Hello mclane,
m> From: mcl...@prima.ruhr.de (mclane)
m> Subject: Re: NIMZO 3
m> Organization: Prima e.V. Dortmund
m> >Hmm, I'd like to see results from MCP6 against Hiarcs5 and Nimzo3.5,
m> We are working on this....
OK, I can wait. :-)
m> >especially with tournament-conditions, not only 60moves/60mins.
m> we are even more working on that. But - as you know - it takes some
m> hours even for my autoplayers.
I am curious about your results.
In our chessnet MCP6 disappoints very much, he is in very bad position
against Rebel8 with black and white and has a lost position against the
King 2.54. Amazing. Is this a Pentium-optmized-program or does he gain so
much strength from the permanent-brain? I cannot believe the results so
far. Of course I will try again for I hope it is only statistic fault..
:-)
m> >I'd also like to see MCP6 play against Wchess which seems to be very
m> >strong, at least in our German Chessnet when playing via e-mail, and
m> >against The King 2.54. Can anybody effort to get results maybe from
m> >autoplayer?
m> I would like to, but my last token of Wchess is on my 486-DX4 100 Mhz
m> PC. Because the token is not uninstallable, and I don't want to buy
m> the program twice, .... shit copy protection. I have nothing against
m> copy protection. But if the tokens are not uninstallable and when I
m> have no support from a dealer...
Very nice. :-(
Why don't we all "boykottiern" this shit copy-protction? Are we so stupid
and/or so eager on having such programs??
Harald
--
mclane
Harald...@p21.f2.n1.z1001.fidonet.org (Harald Faber) wrote:
>I am curious about your results.
>In our chessnet MCP6 disappoints very much, he is in very bad position
>against Rebel8 with black and white and has a lost position against the
>King 2.54. Amazing. Is this a Pentium-optmized-program or does he gain so
>much strength from the permanent-brain? I cannot believe the results so
>far. Of course I will try again for I hope it is only statistic fault..
>:-)
Your CHESS NET ?!
What is that, is that this strange thing Computer-Schach and Spiele
reported in one edition ( not an attack here !! I just have not read
the article).
Can you explain to me ?!
>m> >I'd also like to see MCP6 play against Wchess which seems to be very
>m> >strong, at least in our German Chessnet when playing via e-mail, and
>m> >against The King 2.54. Can anybody effort to get results maybe from
>m> >autoplayer?
German Chess-Net ?!
Isn't it cheaper to send email-games using INTERNET?
Why use Chess-Net. What is Chess-net ?
>m> I would like to, but my last token of Wchess is on my 486-DX4 100 Mhz
>m> PC. Because the token is not uninstallable, and I don't want to buy
>m> the program twice, .... shit copy protection. I have nothing against
>m> copy protection. But if the tokens are not uninstallable and when I
>m> have no support from a dealer...
>Very nice. :-(
>Why don't we all "boykottiern" this shit copy-protction? Are we so stupid
>and/or so eager on having such programs??
I am a Kittinger fan. The temptation is too big.
I would by the latest Kittinger program when it would run on a
Commodore 64 again.
I still wish my old Superconny back. Although I think this is all pure
nostagique stuff.
>Harald
>--
Tord Kallqvist Romstad
mclane (mcl...@prima.ruhr.de) wrote:
: I still wish my old Superconny back. Although I think this is all pure
: nostagique stuff.
Perhaps Kittinger still has the source code for Super Constellation?
I guess a PC version of Super Constellation would sell rather well...
Tord
: >Harald
: >--
Komputer Korner
I too am nostalgic for the Super Conny. If my memory serves me correct
wasn't the Super Constellation a preprocessor in the mold of Nimzo?
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
kouldn't find the real Motive and variation tree in
Harald Faber
Hello mclane,
m> >I am curious about your results.
m> >In our chessnet MCP6 disappoints very much, he is in very bad position
m> >against Rebel8 with black and white and has a lost position against the
m> >King 2.54. Amazing. Is this a Pentium-optmized-program or does he gain so
m> >much strength from the permanent-brain? I cannot believe the results so
m> >far. Of course I will try again for I hope it is only statistic fault..
m> >:-)
m> Your CHESS NET ?!
Not mine, but OUR. ;-)
m> What is that, is that this strange thing Computer-Schach and Spiele
m> reported in one edition ( not an attack here !! I just have not read
m> the article).
No, you remember the GAMENET or so.
I thought to have posted here some information about our chessnet. For it
didn't appear here I send you again the information to your e-mail-adress.
m> Can you explain to me ?!
I'll send you some info.
m> >m> >I'd also like to see MCP6 play against Wchess which seems to be very
m> >m> >strong, at least in our German Chessnet when playing via e-mail, and
m> >m> >against The King 2.54. Can anybody effort to get results maybe from
m> >m> >autoplayer?
m>
m> German Chess-Net ?!
Yes again...
It is a kind of FIDO-net..
m> Isn't it cheaper to send email-games using INTERNET?
No, for I call in the 200km-region for chessnet and >200km for Internet-
newsgroup. It also doesn't last so long to send and receive answers for I
don't get newsgroups-mails every day but in chessnet.
In chessnet you can play public, ask public questions etc and also have
the possibility to do it by private mail. Remember in Geman chessnet we
almost always play offline! So it doesn't make sense to use the slower way
internet.
Tell me where you live and I give you the nearest mailbox-number where you
can get chessnet.
m> Why use Chess-Net. What is Chess-net ?
Again I'll send you info for it didn't appear here. :-(
m> >Why don't we all boykot this shit copy-protction? Are we so stupid
m> >and/or so eager on having such programs??
m> I am a Kittinger fan. The temptation is too big.
But how fast can one lose one or two installations?
m> I would by the latest Kittinger program when it would run on a
m> Commodore 64 again.
Hmm, seems to me that I should try and test WChess, but te c.p. refuses me
from buying.
m> I still wish my old Superconny back. Although I think this is all pure
m> nostagique stuff.
Probably. :-)
Harald
--
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<32DCA...@netcom.ca>...
> Tord Kallqvist Romstad wrote:
> >
> > mclane (mcl...@prima.ruhr.de) wrote:
> > : I still wish my old Superconny back. Although I think this is all
pure
> > : nostagique stuff.
> >
> > Perhaps Kittinger still has the source code for Super Constellation?
> > I guess a PC version of Super Constellation would sell rather well...
> >
> > Tord
> >
> > : >Harald
> > : >--
>
> I too am nostalgic for the Super Conny. If my memory serves me correct
> wasn't the Super Constellation a preprocessor in the mold of Nimzo?
I challenge this statement on the following fronts:
1. Your memory never serves you correct.
2. The Super Constellation is several years old.
3. Six weeks ago we explained what pre-processing was to you.
4. Six weeks ago you didn't know about pre-processing.
5. How could your memory from several years ago contain data only inputted
several weeks ago ?
Chris Whittington
Ed Schröder
"Chris Whittington" <chr...@demon.co.uk> wrote:
http://www.demon.co.uk/oxford-soft
Nice home page!
Well done Chris!
Tip of the week!
- Ed -
>Chris Whittington
Komputer Korner
Chris Whittington wrote:
>
snipped
> I challenge this statement on the following fronts:
>
> 1. Your memory never serves you correct.
>
> 2. The Super Constellation is several years old.
>
> 3. Six weeks ago we explained what pre-processing was to you.
>
> 4. Six weeks ago you didn't know about pre-processing.
>
> 5. How could your memory from several years ago contain data only inputted
> several weeks ago ?
>
> Chris Whittington
>
> > --
> > Komputer Korner
> >
> > The komputer that kouldn't keep a password safe from
> > prying eyes, kouldn't kompute the square root of 36^n,
> > kouldn't find the real Motive and variation tree in
> > ChessBase and missed the real learning feature of Nimzo.
> >
I don't know where you get the idea that I never knew about pre
processing.
What I didn't know was that Nimzo was a preprocessor.
Peter W. Gillgasch
Chris Whittington <chr...@demon.co.uk> wrote:
> Komputer Korner <kor...@netcom.ca> wrote in article
> <32DCA...@netcom.ca>...
[snip]
> > I too am nostalgic for the Super Conny. If my memory serves me correct
> > wasn't the Super Constellation a preprocessor in the mold of Nimzo?
>
> I challenge this statement on the following fronts:
>
> 1. Your memory never serves you correct.
Hmpf.
> 2. The Super Constellation is several years old.
True.
> 3. Six weeks ago we explained what pre-processing was to you.
True.
> 4. Six weeks ago you didn't know about pre-processing.
I don't know what the Komputer knows. You don't know what the Komputer
knows. I am not sure if the Komputer knows what the Komputer knows.
Hell, I am not even sure what I know.
> 5. How could your memory from several years ago contain data only inputted
> several weeks ago ?
Sorry Khris, the Komputer is rikht this time :)
Kittinger does preprocessors and the Constellation was his first.
Thorsten's beloved Virtual is one too I think. Still it plays nice
chess...
-- Peter
mclane
Komputer Korner <kor...@netcom.ca> wrote:
>I too am nostalgic for the Super Conny. If my memory serves me correct
>wasn't the Super Constellation a preprocessor in the mold of Nimzo?
>--
>Komputer Korner
Superconny was ablte to play chess. It sacrified pieces or pawns, had
knowledge and made chess a spectacular event.
You cannot relate to Nimzo....
Sorry....
Komputer Korner
Well then your objection to Nimzo is not because of it being a
preprocessor.
If you liked the Super Constellation as much as I did, then your
objection
to Nimzo must be based on it's perceived lack of knowledge in relation
to other
programs and not because of it being a preprocessor or else you couldn't
love the Super Connie.
Tom C. Kerrigan
Chris Whittington (chr...@demon.co.uk) wrote:
> 3. Six weeks ago we explained what pre-processing was to you.
Bummer that I wasn't here six weeks ago. I have no idea what a
preprocessor is. Makes me feel a bit stupid now.
My guess is that it's a function that sets values for piece/square tables
that's called before the search. Close?
Cheers,
Tom
Peter W. Gillgasch
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote:
> Bummer that I wasn't here six weeks ago. I have no idea what a
> preprocessor is. Makes me feel a bit stupid now.
>
> My guess is that it's a function that sets values for piece/square tables
> that's called before the search. Close?
Close. Anything that modifies the eval based upon the current root
position, piece/square tables being the most obvious example.
-- Peter
May God grant me the serenity to accept the things I cannot change,
courage to choke the living shit out of those who piss me off,
and wisdom to know where I should hide the bodies...
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote in article
<5brftl$b...@merlin.pn.org>...
> Chris Whittington (chr...@demon.co.uk) wrote:
>
> > 3. Six weeks ago we explained what pre-processing was to you.
>
> Bummer that I wasn't here six weeks ago. I have no idea what a
> preprocessor is. Makes me feel a bit stupid now.
>
> My guess is that it's a function that sets values for piece/square tables
> that's called before the search. Close?
Exactly, if only KK was so quick :)
We posted much about how the assumptions made at ply 0 becamse increasingly
off target with increasing search depth.
Chris Whittington
>
> Cheers,
> Tom
>
Komputer Korner
Tom C. Kerrigan wrote:
>
> Chris Whittington (chr...@demon.co.uk) wrote:
>
> > 3. Six weeks ago we explained what pre-processing was to you.
>
> Bummer that I wasn't here six weeks ago. I have no idea what a
> preprocessor is. Makes me feel a bit stupid now.
>
> My guess is that it's a function that sets values for piece/square tables
> that's called before the search. Close?
>
> Tom
No, it is a program that uses almost all its knowledge evaluation at the
root instead of at the leaf nodes. Ex: Fritz and Nimzo are preprocessors
and Rebel, Hiarcs, and M-Chess are leaf node processors. So far leaf
node
processors have always done better.
mclane
"Chris Whittington" <chr...@demon.co.uk> wrote:
>--
>Exactly, if only KK was so quick :)
>We posted much about how the assumptions made at ply 0 becamse increasingly
>off target with increasing search depth.
>Chris Whittington
>>
>> Cheers,
>> Tom
>>
OFF TARGET !
Ply 0 decisions:
Example:
He is a male. So he will be stong and I can mary him.
Marriage happens. Also 3 years.
Female finds out: he is a male, but a very macho-man. Not able to
stand anything nor to lose anything.
The marriage gets divorced.
Ply 0 decisions are fast, but often wrong.
mclane
Komputer Korner <kor...@netcom.ca> wrote:
>Tom C. Kerrigan wrote:
>>
>> Chris Whittington (chr...@demon.co.uk) wrote:
>>
>> > 3. Six weeks ago we explained what pre-processing was to you.
>>
>> Bummer that I wasn't here six weeks ago. I have no idea what a
>> preprocessor is. Makes me feel a bit stupid now.
>>
>> My guess is that it's a function that sets values for piece/square tables
>> that's called before the search. Close?
>>
>> Cheers,
>> Tom
>No, it is a program that uses almost all its knowledge evaluation at the
>root instead of at the leaf nodes. Ex: Fritz and Nimzo are preprocessors
>and Rebel, Hiarcs, and M-Chess are leaf node processors. So far leaf
>node
>processors have always done better.
And will always do.....