Opening Books - Something does not add up

[spacious_mind]

I would imagine there is essentially one book but within it it has moves assigned to various book styles. Some moves may be exclusive to a book style while others may occur in different styles.

That seems the most likely implementation to me.

I am starting a separate Post on a subject which I still struggle with matching up to certain numbers provided in different ways:

First of all before I start composing my questions, the above comment from Mark I do believe is a Key to part of the solution.

If you look into the Cosmos Manual the following is described:

2.7 Built-in Openings
At the beginning of a game, the computer will often move instantly on
many levels. This is because it is playing from memory, using its own
built-in "book" of opening chess moves. This book contains thousands
of positions, including most major openings and many positions from
grandmaster play. If the current board position is in its book, the computer
plays a response to that position automatically, instead of having
to think about the move!
A special feature of this computer's opening book is its ability to
handle transpositions. A transposition occurs when a position reached
by a certain set of moves can also be reached when those same moves
occur in a different order. The computer's integrated Automatic Transposition
Manager handles these cases with ease!
Also included is a unique user-selectable book feature, which lets
you choose from different types of opening books, or even turn the
book off, if desired.

e. Full Book (Square E2)
Turning on the Full Book option gives the computer the freedom to
choose any move from its built-in book of openings, so you'll see a
greater variety of opening lines being played. If this option is selected
the Passive, Active, and Tournament Book options are disabled.
With this option on, you may see the computer make some questionable
moves. This is because its built-in book of openings must contain
responses to certain lines of play (even questionable lines), in case they
are played. While the computer would not make these moves on its
own, it needs to know how to respond to them in the best way. Therefore,
when you turn on the Full Book, the computer could potentially
play one of these moves.

f. Passive Book (Square F2)
Selecting the Passive Book option forces the computer to give
preference to passive and closed positions when deciding which lines to
play.If this option is selected, the Active Book and Tournament Book options are disabled.

g. Active Book (Square G2)
When this option is activated, the computer will give preference to
active lines and open positions when deciding which lines to play.
If this option is selected, the Passive Book and Tournament Book options are
disabled.

h. Tournament Book (Square H2)
When you turn the Tournament Book option on, the computer is
forced to always select the best possible line of play in every opening.
While this results in the best chess play, it also narrows down the
computer's choice of moves by limiting its available book lines. If this
option is selected, the Active Book and Passive Book options are
disabled.

If you read the above carefully it seems to fit into Mark's comment that only one book should be used.

Another hint that this seems to be correct comes from Mephisto Modena:

Mephisto's Opening Book differentiates between active and passive moves. Active opening moves are the ones the computer will select when it is playing. Passive moves however are recognized by MEPHISTO so that it can find the answering moves if the opponent plays these but it will avoid playing them itself as it could lead to positions which the computer's program may not handle so well. Passive moves are indicated by dots between the characters and numbers on the display.

Now all this nicely fits into Mark's explanation.

Now here are some programs and their ROM sizes and the year that they were first sold:

Mephisto Modena - 1992
ROM 32,768 Bytes
Opening Book: 10,000 Half Moves
Processor: 6502 - 4 MHz

GK2000 - 1992
ROM: 16,384 Bytes
Opening Book: 2,000 Half Moves
Processor: H8 - 10 MHz

GK2100 - 1993
ROM 32,768 Bytes
Opening Book: 30,000 Half Moves
Processor: H8 - 10 MHz

Saitek Travel Champion 2100 - 1994
ROM 32,768 Bytes
Opening Book: 30,000 Half Moves
Processor: H8 - 10 MHz

RadioShack Champion 2250XL - 1997
ROM 32,768 Bytes
Opening Book: 20,000 Half Moves
Processor: H8 - 10 MHz

Saitek Cosmos - 1998
ROM 32,768 Bytes
Opening Book: 6,000 Half Moves
Processor: H8 - 10 MHz

Saitek Cougar - 1998
ROM 32,768 Bytes
Opening Book: 6,000 Half Moves
Processor: H8 - 16 MHz

Mephisto Milano Pro - 1997
ROM: 65,536 Bytes
Opening Book: 50,000 Half Moves
Processor: SuperH 7034 20 MHz

Mephisto Magellan - 1998
ROM: 65,536 Bytes
Opening Book: 50,000 Half Moves
Processor: SH7034 20 MHz

Mephisto Atlanta - 1997
ROM: 65,536 Bytes
Opening Book: 50,000 Half Moves
Processor: SH7034 20 MHz

Mephisto Senator - 1998
ROM: 65,536 Bytes
Opening Book: 50,000 Half Moves
Processor: SuperH 7034 20 MHz

Here is some data I have taken from DOS Programs:

Fritz 1 - 1991
Fritz.bin 28,828 08/27/91
Fritz.fbk 16,132 08/16/90
-------------------------
Total = 54,960 Bytes

Fritz.exe 233,572 10/07/91

Fritz 2 - 1993
Fritz2.bin 36,037 01/08/93
Fritz2.fbk 22,670 01/13/93
--------------------------
Total = 58,707 Bytes

Fritz2.exe 244,858 12/09/92

Fritz 3 - 1995
fritz3.bin 53,990 01/23/95
fritz3.fbk 140,414 11/28/94
--------------------------
Total = 194,404 Bytes

fritz3.exe 258,428 10/12/95

Mychess - 1980
Mychess.com 52,284 01-01-80

Gideon - 1993
Gideon.trl 29,181 Bytes
Gideon.mvs 42,206 Bytes

So now questions, questions... and more questions..

GK2000
ROM: 16,384 Bytes
Opening Book: 2,000 Half Moves

If you look at the ROM size, I can only assume that the program is a pre Fritz 1 version and pruned down to it's barest minimum. The Opening Book size as shown in Wiki seems to make sense in this case.
You have to also consider that besides the book and the program you also need some sort of minimum basic programming to steer the chess program and features and options.

Mephisto Modena - 1992
ROM 32,768 Bytes
Opening Book: 10,000 Half Moves

Now if you take Fritz.bin 28,828 Bytes this could theoretically fit with a little pruning down of Fritz 1 (which was written in 1991)
But now the question for you programmers, how many bytes do you need for 10,000 Half Moves? Fritz 1 has a book size of 16,132 Bytes. Does anyone know how many Half Moves the Fritz 1 book has. The FBK file is already written in I would guess Assembly language and therefore is already most likely compressed. (I know I cannot read it when I open the file in the DOS Edit mode)

GK2100 - 1993
ROM 32,768 Bytes
Opening Book: 30,000 Half Moves

So how do you fit 30,000 Half Moves into the GK2100?

RadioShack Champion 2250XL - 1997
ROM 32,768 Bytes
Opening Book: 20,000 Half Moves

How do you fit 20,000 into the RS2250XL ? Come on you Programmers Fess UP!!!

Saitek Cosmos - 1998
ROM 32,768 Bytes
Opening Book: 6,000 Half Moves

Saitek Cougar - 1998
ROM 32,768 Bytes
Opening Book: 6,000 Half Moves

Interesting that Wiki shows Cosmos and Cougar as 6,000 Half Moves, as well as Centurion and some others. So what is correct?

Mephisto Atlanta - 1997
ROM: 65,536 Bytes
Opening Book: 50,000 Half Moves

Now here finally!!! I see light with Fritz1 and Fritz2 both fit comfortably into Atlanta. How you may ask?

Ok a theory!

Assume the manuals do not tell the whole truth. Which of course we all know is an easy thing to assume.

Now take a claim of 50,000 Half Moves and Mark's comments and think them through.

What do we have?
Big Book
Tournament Book
Active Book
Passive Book

Now lets do some assumptions.

Big Book = 20,000 Half Moves
Tournament Book = 12,000 Half Moves
Active Book = 9,000 Half Moves
Passive Book = 9,000 Half Moves

What do we have?

We have a book size of 20,000 Half Moves or for arguments sake 20K or 20,480 Bytes in size. (Taking Mark's comment to heart here)
Now look at Fritz 2 again:

Fritz 2 - 1993
Fritz2.bin 36,037 01/08/93
Fritz2.fbk 22,670 01/13/93
--------------------------
Total = 58,707 Bytes

Now I would think that with Atlanta's ROM: 65,536 Bytes Fritz 2 would fit in there nicely and the book, with room to spare for Features and all the other bells and whistles. (Now if someone states that Atlanta is a Fritz 3 then we have a bit of a pruning problem or book size problem again)

So now back to GK2100 and RS2250 how do you fit a book size of 30,000 and 20,000 half moves into 32 KB ROM?

Come on let's hear the explanations!

Currently I preferring to lean on what I read for Modena's book size and even Cosmos, Cougar's books size.

Still I hate all these assumptions that I am making it would be nice if someone had some numbers on how many Bytes you need for 1000 Half Moves!

Interesting stuff regards,

[Theo]

Nick,


you're comparing the size of Assembler code for x86 with code for 6502. I don't think it is quite appropriate. Besides, Fritz1 and and Fritz2 had also quite some graphics and menues. And maybe also databases?!

My thoughts on this are: 2000 opening moves for the GK2000 with its 16kB sounds true. 10.000 moves for the Modena with 32kB also sounds true.

What I don't understand is why some 32kB Morsch clones have a 30.000 library, while others have only 6.000?!

30.000 moves in 32.000 bytes means there is a maximum of 1 byte per move available?!

I can only assume they are stored in ChessBase style: you do not store the "from" and the "to" field, rather you store the position of the move in the list generated by the move generator.

There are 20 moves in the original position, lets say you want 1.Nc3 to be in your library. It is the 18. move in the list. So you only store the number "18". That saves you a lot of space.

In the opening positions there shouldn't be more than 64 possible moves. Thats 6 bits needed for storing (2^6 = 64). But then you need extra Bits to somehow connect those stored moves?! Sounds like you need 1 Byte per move?!

And thus, if 32kB contain 30.000 opening moves, is there space for the actual program & interface left ?!

Maybe some programmer can reveal the truth?!


Speculating Regards,
Theo

[spacious_mind]

Nick,


you're comparing the size of Assembler code for x86 with code for 6502. I don't think it is quite appropriate. Besides, Fritz1 and and Fritz2 had also quite some graphics and menues. And maybe also databases?!

My thoughts on this are: 2000 opening moves for the GK2000 with its 16kB sounds true. 10.000 moves for the Modena with 32kB also sounds true.

What I don't understand is why some 32kB Morsch clones have a 30.000 library, while others have only 6.000?!

30.000 moves in 32.000 bytes means there is a maximum of 1 byte per move available?!

I can only assume they are stored in ChessBase style: you do not store the "from" and the "to" field, rather you store the position of the move in the list generated by the move generator.

There are 20 moves in the original position, lets say you want 1.Nc3 to be in your library. It is the 18. move in the list. So you only store the number "18". That saves you a lot of space.

In the opening positions there shouldn't be more than 64 possible moves. Thats 6 bits needed for storing (2^6 = 64). But then you need extra Bits to somehow connect those stored moves?! Sounds like you need 1 Byte per move?!

And thus, if 32kB contain 30.000 opening moves, is there space for the actual program & interface left ?!

Maybe some programmer can reveal the truth?!


Speculating Regards,
Theo

You are not looking at it correctly the menus are in the exe file which I showed separately.

Again I challenge you all to provide the size for 1,000 Half Moves and then lets calculate.

regards

[chessguru]

Hallo Nick,

Interesting that Wiki shows Cosmos and Cougar as 6,000 Half Moves, as well as Centurion and some others. So what is correct?

the information comes from Saitek:

• Saitek Prospekt 1998 HD

• Saitek Prospekt 1998

Gruß
Micha

[Steve B]

Hallo Nick,

the information comes from Saitek:

Gruß
Micha

Hi Micha
and the information from the Mephisto/Saitek brochures are that the Atlanta and Magellan are the same computers,same specs ,same estimated elo's
nice to see someone actually quoting the written literature one in awhile

Regards
Steve

[Sargon1972]

Hi Guys,

Yes for me it was always clear that Atlanta and Magellan where the same.

Mostly when they launch a "new"program in Modules, there is also mostly a plastic unit with the same program

The Saitek Brochure spoken regards,Hans

[spacious_mind]

For me nothing is clear, and neither is it for you unless you choose to believe that it is clear. But that's your choice and it is another one of the many things that makes our hobby so interesting.

Why is nothing clear?

Well lets see... How about starting with Mach II 4 or 5 Versions of which only one is clearly advertised (L.A. Version) For the rest I guess you take pot luck if it is a 2a, 2b, 2c.... L.A.

How about some other Fidelities?
And I am sure that there is other Fidelities that could be added to this. So what do the Brochures say?

Novag? Versions A, B & C popping up left right and center and I am sure that brochures show all that as well correctly right?

Kaplans? Multiple Versions all advertised? I think not.

So how are all these differences found out? Through playing the games right? And of course in many cases opening them up.

Did Tasc advertise that there are multiple SB30 Versions? I think not.

Well anyway these are just some examples.

The fact remains that Atlanta has more deviations in it's play then a normal clone.. And the ELO's reflect this. How about all of you owners pulling out your Atlanta's, Magellan's, sit down and spend a couple of hours replaying the 6 tests games and show us what yours do! Let's all of us get to the bottom of this!

Now lets talk about the books... How can President, GK2100 and TC2100 have 30,000 opening moves? And a Cosmos, Cougar and Centurion show 6? MMVI 35,000? RS2250 XL 20,000 Moves ? Where does this come from, after all basically they all have to fit into the same ROM and they are all the same programs? So what is wrong with this picture

OK Let's look at Cosmos:

Saitek Cosmos - 1998
ROM 32,768 Bytes
Opening Book: 6,000 Half Moves
Processor: H8 - 10 MHz

6,000 Half Moves? Now to me that makes much more sense after all GK2000 is 2,000 Half Moves which needed to be fitted into into 16 KB.

Well there is no question that Cosmos and Company are better programs. So we have to assume that the program is an improvement after all both play on a H8 10 MHz. So it can't be the speed? Maybe the Manufacturer added a 0 to his advertising with some of the others after all who can prove otherwise? None of us are going to be able to look inside and count it?

Also why is it exactly 6,000, 20,000, 30,000, 35,000 Who counts this ? LOL

Why not be precise if you are going to believe an advert. 19, 743 for example. Now that I would call accurate!

Ok so back to the Cosmos theory!

Fritz2 = 1993. So that one is impossible since GK2100 came out in 1992 and Cosmos is a clone of GK2100!

So lets look at Fritz1 = 1991.

Fritz 1 - 1991
Fritz.bin 28,828 08/27/91
Fritz.fbk 16,132 08/16/90
-------------------------
Total = 54,960 Bytes

Theory:

Big Book = 40% of 6000 = 2,400 (someone would need to show me that these books are 10 times bigger than a GK2000)
Tournament Book = 25% = 1,500
Active Book = 20% = 1,200
Passive Book = 20% = 1,200

= Total 6,300 Half Moves but book size = 2,400 Half Moves (perhaps 20% bigger than GK2000)

So now we have Fritz 1 inside Cosmos as for example 28,828 Bytes plus Book 2,458 Bytes = 31,286 and this leaves a tiny bit of space for the Bells and Whistles.

I know this all sounds so outrageous, but how can this be any less outrageous than believing as Gospel what you are reading in manuals and Brochures ??

This is a great topic and sure I am behaving tongue in cheek so please enjoy and throw in your inputs and theories !!

Perhaps one of us ends up hitting the nail on the head!

Best regards,

Nick

[Steve B]

Well lets see... How about starting with Mach II 4 or 5 Versions of which only one is clearly advertised (L.A. Version) For the rest I guess you take pot luck if it is a 2a, 2b, 2c.... L.A.

Novag? Versions A, B & C popping up left right and center and I am sure that brochures show all that as well correctly right?
So how are all these differences found out? Through playing the games right? And of course in many cases opening them up.

Did Tasc advertise that there are multiple SB30 Versions? I think not.

Hi Nick
Well i can answer here

Computer Chess Reports wrote extensively about the different Mach IIA B C and LA versions

the Novag Super Experts/Super Fortes A B and C were well advertised when released for sale
i have brochures on each version
in addition the new version was shown on the box and an addendum was included with the computer explaining the differences between the new version and the old
not to mention extensive coverage in Computer Chess Reports

nobody has to play even one game to know this information

As to the two different SB 30 Boards
yes i knew about it a long time ago and posted about it in Kurt's old site
i have a board that works with the 2.2 and one that does not
a dutch collector that sold me one of the boards included an article about this but i no longer have it so it was written about at least in Dutch
what we did not know about was the Gideon eprom

its one thing to have a healthy skepticism about the hobby's rich literature but it is quite another to just disregard it and try to reinvent the wheel all over again

you can believe the RS has only 6k books.Atlanta is a different computer then Magellan and Geller is a stronger player then Fischer
i can believe RS probably has 20k books,Atlanta and Magellan are so close they are for all intent and purposes ..clones.. and Fischer eats Gellers lunch

Viva La Difference Regards
Steve

[spacious_mind]

Novag? Versions A, B & C popping up left right and center and I am sure that brochures show all that as well correctly right?




So how are all these differences found out? Through playing the games right? And of course in many cases opening them up.

Did Tasc advertise that there are multiple SB30 Versions? I think not.

Hi Nick
Well i can answer here
the Novag Super Experts/Super Fortes A B and C were well advertised when released for sale
i have brochures on each version
in addition the new version was shown on the box and an addendum was included with the computer explaining the differences between the new version and the old
not too mention extensive coverage in Computer Chess Reports
nobody has to play even one game to know this information

As to the two different SB 30 Boards
yes i knew about it a long time ago and posted about it in Kurt's old site
i have a board that works with the 2.2 and one that does not
a dutch collector that sold me one of the boards included an article about this but i no longer have it
what we did not know about was the Gideon eprom

its one thing to have a healthy skepticism about the hobby's rich literature but it is quite another to just disregard it and try to reinvent the wheel all over again

you can believe the RS has only 6k books.Atlanta is a different computer then Magellan and Geller is a stronger player then Fischer
i can believe RS probably has 20k books,Atlanta and Magellan are so close they are for all intent and purposes ..clones.. and Fischer eats Gellers lunch

Viva La Difference Regards
Steve

Hi Steve,

Yes, I know I threw in a couple of examples which may not all be perfect, but I think both you and I know there are enough other examples that we could come up with that prove multiple rom versions exist inside computers without a manufacturer telling you this.

Also there is time difference between Atlanta 1997 and Magellan 1998. So I do not see this as cast in stone that everyone of these has the same ROM version or have a small program adjustment.

Not asking you to believe me that's the whole point of this post, lets hear what everyone has to say about these inconsistencies.

It would be nice however if you could play the test and explain the move variations between Atlanta and Magellan. Especially since there are enough examples of 100% clone behaviors supported by different Testers, which makes it kind of hard to disprove the validity of these tests.

Anyway let's enjoy our different opinions.. Besides you forgot to write about Mach II in your above post

Best regards

[Steve B]

Included the Mach II
you always get me in between edits!


i did not try your tests because as i mentioned before i do not believe tests with pondering ON are useful for clone detection
what i did take the time to do was in that thread 7 years ago..mimick your test move by move but with pondering OFF and as i showed then i had not one difference in any of the moves
i started the other thread regarding the Atlanta/Magellan issue and your new game tests to find out... IF your own tests showed a high correlation would you then admit they were the same?
i even agreed that if there were a high mismatch then they cannot be clones
obviously if they do not at the end show a high correlation then i dont expect you to admit that
but the question i was interested in with that new thread was to see if you would agree they were the same even if your own tests showed a high match ..


my guess is that even if they do wind up showing a high match you will still not agree and you will point to some other tests made somewhere else or some other games played a week ago last Tuesday and we will still be in the same place


Spinning Wheels Regards
Steve

[Larry]

Not asking you to believe me that's the whole point of this post, lets hear what everyone has to say about these inconsistencies.

Well regarding companies releasing information about their products,
my experience with producers of anything is that they will release
information which will suit their purpose. You will often see in the fine
print a note about technical change being made in the interests of progress, or something similar. You will remember the Fidelity SC9, in
which late production run units carried a different program running
at a different speed, but nothing mentioned by the company.
It seems to me that for a lot of manufacturers of anything there may well
be a disconnect between the technical team and the marketing team.
In the case of chess computers, they would not have seen it as a problem,
because after all, how many hobbyists out there will go to the expense of
owning both, eg, an Atlanta *and* a Magellan? Any differences would
go unnoticed for decades, after which time who cares? The head of the
marketing team may live and work in a different city to the technical
team guys, and may not even known them personally. Under orders to
get advertising information out to the retailers ahead of the computer
release date, it would be fairly normal to print at least some of that
information based on verbal data received by someone over the phone.
If we place ourselves in the shoes of the key people involved in those
heady days of chess computers, we can appreciate that time constraints
factored into some decision making.
L

[Steve B]

Well regarding companies releasing information about their products,
my experience with producers of anything is that they will release
information which will suit their purpose.

Thats true and thats why there were publications like Computer Chess Reports starting in 1983 and Selective Search Magazine starting in 1985
which informed buyers and seperated fact from fiction

when i use the term "Literature" i do not mean the Manufacturers literature only.. i also mean books that were written about Dedicated chess computers and also the magazines dedicated to keeping the Manufacturers in line

A Vast Wealth Of Literature Regards
Steve

[spacious_mind]

Included the Mach II
you always get me in between edits!


i did not try your tests because as i mentioned before i do not believe tests with pondering ON are useful for clone detection
what i did take the time to do was in that thread 7 years ago..mimick your test move by move but with pondering OFF and as i showed then i had not one difference in any of the moves
i started the other thread regarding the Atlanta/Magellan issue and your new game tests to ask you ...
IF your own tests showed a high correlation would you then admit they were the same?
obviously if they do not at the end show a high correlation then i dont expect you to admit that
but the question i was asking and the reason for that seperate thread is to establish that ..IF they did show a high match..
would you then admit it?

my guess is that even if they do wind up showing a high match you will still not agree and you will point to some other tests made somewhere else or some other games played a week ago last Tuesday and we will still be in the same place


Spinning Wheels Regards
Steve

Hi Steve,
This is such a hard thing to explain. I absolutely believe that Magellan and Atlanta have the same program and the same I believe with Senator. But of course the Senator can be explained away because it does not have the 512KB ram. So there is a difference between Atlanta/Magellan and Senator.

Now back to Atlanta/Magellan. That in Brute Force mode they are identical I would not disagree with you at all (although I have never actually spend 5 minutes to try this)

But in selective play there are move variations which are almost as big variations as RS2250 and say TC2100. And I cannot explain these variations they are not your normal 95% they are more in the region of 85% - 95% depending on the game. And for this reason all that I am a proponent for is that they are listed separately in Wiki's ELO Lists.

If I was to fall for your argument to bundle them, then I would also since Micha has stated over and over again that in Brute Force RS2250XL and President are identical even going as far as the same move BEEP! I would then have to concede to the possibility of RS2250XL being bundled with the GK2100 clones on these same ELO lists.

This is the Key thing that this discussion is centered around. We don't play Brute Force we all play selective and the ELO ratings reflect that.

So yes Steve I agree that they are the same Program but when we play with them in Selective mode there is a behavior that justifies an ELO List separation for Atlanta and Magellan as well as RS2250XL and GK2100.

Wiki does state both are related. How would you define clearer and at the same time allow them to enjoy their separate ELO Listing?

I think clone is not the perfect word here because the same argument would also apply to RS2250XL.

Maybe something like:
Same Program with some variation in Selective Mode?

Best regards

[spacious_mind]

Well regarding companies releasing information about their products,
my experience with producers of anything is that they will release
information which will suit their purpose.

Thats true and thats why there were publications like Computer Chess Reports starting in 1983 and Selective Search Magazine starting in 1985
which informed buyers and seperated fact from fiction

when i use the term "Literature" i do not mean the Manufacturers literature only.. i also mean books that were written about Dedicated chess computers and also the magazines dedicated to keeping the Manufacturers in line

A Vast Wealth Of Literature Regards
Steve

Well while I may have never pursued Selective Search too much I do have the complete set of Computer Schach & Spiele the German Chess Computer Bible as well the Computer Chess Reports and Modul-Schach-Computer Tests.

I have been rereading them lately a lot and for the most part I agree with your statement but there are occasional errors as well. They did not have these high speed laptop gizmos that we have today, so many of the strength tests are not 100% reliable and have certain biases depending on which brand or programmer you support and like.

Great reading material and knowledge base but not everything you read is 100% as viewed from the year 2013.

Best regards,

[Larry]

Great reading material and knowledge base but not everything you read is 100% as viewed from the year 2013.

+1