Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The importance of opening books -- a simple experiment

Author: Vasik Rajlich

Date: 06:39:17 02/20/05

Go up one level in this thread


On February 20, 2005 at 08:22:53, Sandro Necchi wrote:

>On February 20, 2005 at 07:30:41, Vasik Rajlich wrote:
>
>>On February 20, 2005 at 06:34:30, Sandro Necchi wrote:
>>
>>>On February 20, 2005 at 05:21:35, Vasik Rajlich wrote:
>>>
>>>>On February 20, 2005 at 04:40:19, Sandro Necchi wrote:
>>>>
>>>>>On February 19, 2005 at 07:43:49, Vincent Diepeveen wrote:
>>>>>
>>>>>>On February 17, 2005 at 14:03:30, Tord Romstad wrote:
>>>>>
>>>>>Hi Vincent,
>>>>>
>>>>>I hope everything is well with you. I guess you'll meet Stefan at Paderborn.
>>>>>
>>>>>>
>>>>>>Don't do idiot experiments that just support the idiocy you invented yourself.
>>>>>>
>>>>>>IF you have an engine rated 3000 in strength THEN it will play virtually achieve
>>>>>>2300 in the important games instead of 3000 when NOT using a book. So the book
>>>>>>delivers 700 rating points.
>>>>>
>>>>>This is a very interesting statement.
>>>>>I agree with you and I know that the stronger an engine is the best it will get
>>>>>from a good opening book, but I never estimated how much it would get reaching
>>>>>3000 in strenght.
>>>>>
>>>>>>
>>>>>>Experiments with some idiot engine that is itself rated what is it, 2200,
>>>>>>will be useless of course.
>>>>>>
>>>>>>Some years ago when engines were 2200 level it was the Mchess-Necchi team who
>>>>>>said a book was worth 300, and he meant that obviously in the same way as i mean
>>>>>>the 700 points at real high level.
>>>>>
>>>>>Yes, you are correct.
>>>>
>>>>So let me get this straight - if two equal engines play, and one of them has a
>>>>book that decides one half of all games in its favor without any chance for the
>>>>victim to continue, this books needs to be improved, since the rating boost here
>>>>is only 200 rating points.
>>>
>>>A good book, to me should do the following:
>>>
>>>1. Give positions with a plus for the program.
>>>2. Give positions which can show the best of the program strenght by allowing it
>>>to play the strongest.
>>>3. Play positions that bring to endgames the program know to play well.
>>>
>>>Now if 2 engines about strong the same reach a position which is better for
>>>program a (suitable more for program a) program a will get an advantage soon or
>>>later even if not higher in score from the opening and program b will not be
>>>able to recover.
>>>
>>>The result is that in that game program a and b will not show to be about strong
>>>the same. Do you understand what I mean?
>>
>>Sandro,
>>
>>yes, I understand what you say. Certainly, you can benefit from having a book
>>which co-operates with your engine. I can believe that when you have an engine
>>at the level of Shredder, every improvement helps.
>>
>>The question we're dealing with here is - how much does it help?
>
>It depends on the program strenght, in how good is the book for that specific
>program and where we are using it.
>
>According to Vincent if a program is 3000 points strong the very good book helps
>700 points.
>
>When programs where 2200 points strong a very good book (for that program)
>helped 300 points.
>
>We are talking about WCCC tournament, so with selection guided by the operator
>and a few games.
>
>
>>
>>Personally, I doubt that you can regularly patch up problems in the evaluation
>>via book. Or that you can regularly exploit positional problems in your
>>opponents via book. Or that you can just get forced wins in the opening on a
>>regular basis.
>
>I did not say that.
>We are studying our opponents like true chess players and make estimation of
>what we can do better, worst and so on...
>The book is optimized on these things based on our program knowledge and not on
>others weaknesses...
>
>>
>>Of course, this is just a personal opinion.
>>
>>The question which Tord and Uri and Martin and Sune and now also me are asking
>>is: what experiments support the claim that the book has a significant (ie. >=75
>>Elo) effect on the engine level?
>
>A book made from games database on statistics has very little influence as good
>things are compensated negatively by bad things...and there is no relation
>between the player to whom the book is given and the moves selection.
>It is like to force you to play what Kasparov does without you being able to
>play like Kasparov...(maybe you play better...I do not know...)
>

Sandro -

I pretty much agree with everything you say - at least, everything but the
rating estimates. Of course that's pretty much the summary of this entire huge
thread :)

Anyway, if your rating estimates are anywhere near correct, then chess sucks and
we should all switch to Fischer random :)

Long live the middlegame!

Vas

>>
>>The answer that you must try playing in a WCCC with a top engine but without a
>>book is not very satisfying :)
>
>Without a book is a disaster...without a good book...time wasted if you play to
>win of course...
>>
>>Thanks,
>>Vas
>
>Sandro
>>
>>>
>>>>
>>>>Some years ago I played something like 300-400 games is tournaments. From all
>>>>these games, maybe 3 or 4 were decided in the opening - at least as far as I
>>>>could tell. So I'm wondering what exactly I missed there ...
>>>
>>>In WCCC tournaments it is different as the operator is choosing the variations
>>>and these are very selected. The selection is made also on the weakness of the
>>>opponent and not only in our strenght.
>>>
>>>This is the difference of a hand made book compared to a book made from games
>>>database. This would take about 1 hour or so, while I have been working on my
>>>book since 27 years now and making improvements every day...
>>>
>>>>
>>>>Vas
>>>
>>>Sandro
>>>
>>>>
>>>>>
>>>>>>
>>>>>>The weakest link is what counts.
>>>>>>
>>>>>>I don't care for your engine in that respect, let alone such stupid experiment.
>>>>>>
>>>>>>Take a strong engine from which you feel it is the best engine in the world.
>>>>>>Show up without book in important events. World champ will be the best test.
>>>>>>
>>>>>>THEN calculate after a 100 years of doing that, what the odds were you won that
>>>>>>event. You will see it's 0%.
>>>>>
>>>>>Yes, here we agree too.
>>>>>
>>>>>Sandro
>>>>>>
>>>>>>750 rating points is roughly meaning that a person A has 0% chance against
>>>>>>person B.
>>>>>>
>>>>>>What i'm saying is that not using a book versus an excellent book is making that
>>>>>>difference true. 700 rating points. Not a point less.
>>>>>>
>>>>>>Vincent
>>>>>>
>>>>>>>A couple of days ago, a well-known programmer and regular
>>>>>>>poster here on the CCC claimed that a good opening book
>>>>>>>was worth at least 700 Elo points.  I thought this number
>>>>>>>looked completely outrageous, and decided to do a simple
>>>>>>>experiment.
>>>>>>>
>>>>>>>I am the author of a basic and minimalistic UCI chess engine
>>>>>>>called Glaurung.  Source code and executables for Mac OS X,
>>>>>>>Linux and Windows can be found at the following URL:
>>>>>>>
>>>>>>>http://www.math.uio.no/~romstad/glaurung/glaurung.html
>>>>>>>
>>>>>>>Recently, I have played some test matches with Glaurung
>>>>>>>against the strongest engine I have on my compter: Hiarcs
>>>>>>>9.6.  Not surprisingly, all such matches end in crushing
>>>>>>>victories for Hiarcs.  The last match I played ended
>>>>>>>75-25 in Hiarcs' favor.
>>>>>>>
>>>>>>>As a crude test of the "good book=700 Elo" claim, I have
>>>>>>>now repeated the match with identical program versions
>>>>>>>and conditions, except that Hiarcs was now playing without
>>>>>>>an opening book.  Assuming that Hiarcs' book is worth 700
>>>>>>>Elo, the expected result of this second match would be
>>>>>>>something like 95-5 in _Glaurung's_ favor.
>>>>>>>
>>>>>>>The actual result of the second match was very close to
>>>>>>>the first match:  Hiarcs won by 72-28.
>>>>>>>
>>>>>>>As far as I can see, this means that at least one of the
>>>>>>>following must be true:
>>>>>>>
>>>>>>>a) The statement "good book=700 Elo" is lightyears away
>>>>>>>from the truth.
>>>>>>>
>>>>>>>b) Hiarcs has an extremely bad opening book, and with a
>>>>>>>half decent opening book it would be several hundred
>>>>>>>rating points ahead of Shredder.
>>>>>>>
>>>>>>>
>>>>>>>Tord



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.