Computer Chess Club Archives


Search

Terms

Messages

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

Author: Vasik Rajlich

Date: 01:58:05 02/24/05

Go up one level in this thread


On February 23, 2005 at 17:35:34, Peter Berger wrote:

>On February 23, 2005 at 16:51:11, Vasik Rajlich wrote:
>
>>On February 23, 2005 at 11:36:03, Peter Berger wrote:
>>
>>>On February 23, 2005 at 06:52:07, Vasik Rajlich wrote:
>>>
>>>>2) Surely there is some difference in level, but how much? I have yet to see
>>>>even one statistic that suggests such a difference for any two equally-matched
>>>>engines. There are plenty of experiments that could show this, so it's at least
>>>>suspicious that it hasn't happened yet.
>>>
>>>That's an interesting idea. Yes, I have not seen such a thing either. I know the
>>>effect is true and can easily be more than 100 ELO when it is about engines of
>>>different level. I don't see a very good reason why this should not work with
>>>engines of similar level, but I have no data to support it.
>>
>>Ok, how about the following experiment:
>>
>>1) You pick two engines of similar level, and post here 25 opening positions
>>where you think engine A will do better, and 25 opening positions where you
>>think engine B will do better.
>>2) I'll run the 100 games on my computers. You won't know in advance the time
>>control I use, or which versions of the two programs I use.
>>
>>Good luck getting 66 points :)
>
>As I said, this is an interesting observation you made and I am not in the
>position to contradict you. It's also kind of a challenge, and I love to gamble,
>so maybe .. :)
>

Ok - if you take up the challenge, you can post your positions on this forum - I
check here pretty regularly.

>But it is possible to get close to 40% of the points with Crafty against a top
>professional this way, both running on single CPUs, contrary to 25-30% with
>default books.
>

Certainly you can get this by choosing drawn positions. :) It would leave fewer
questions if you could show this effect between two top engines, but I guess
other setups are also possible.

>>>>>>IMHO - an amateur engine shouldn't bother with book until let's say four years
>>>>>>go by. At that point, some professional chess player should be hired who will
>>>>>>start from an automatically generated book and spend some month or so making
>>>>>>some adjustments, preferably in the range of moves 5-15. In addition to this, it
>>>>>>may make sense to keep running the engine and "pre-computing" some results,
>>>>>>which can be spot-checked as appropriate. This last step will especially help at
>>>>>>fast time controls.
>>>>>>
>>>>>>Vas
>>>>>>
>>>>>
>>>>>I don't think this approach is optimal , but it will lead to a very reasonable
>>>>>book.
>>>>
>>>>So what is the optimal approach? (Aside from full-time professional chess player
>>>>spending 60 hours a week on the book? :))
>>>>
>>>>Vas
>>>
>>>Please allow me to be vague and add some IMHO at random points ;). If there is
>>>something like an automated book as a starting point, it should be extremely
>>>limitted, so that there is no risk for it to contain blunders. Ideally there
>>>should be no line in the book the author isn't aware of. As you pointed out
>>>yourself, there is a lot of stuff that has to be added manually anyway, e.g.
>>>replies to uncommon openings that can be used to fool the computer ( e.g. all
>>>kinds of unusual gambits). Although they are unlikely to come up in a tournament
>>>they are important. The core of the repertoire should be worked out more
>>>thoroughly than that and profit from existing human analysis, after having been
>>>blunderchecked carefully with program. A few novelties and surprises can't hurt
>>>either.
>>
>>Of course, the more is hand-checked, the better. I just don't know how realistic
>>it is to hand-check everything.
>
>It is possible (some stuff can also be automated) - only sometimes you'll find
>yourself out of book after 1. d4 d5 2. c4 e6 3. cxd5 if you are not careful :)
>

Well - that's life I guess. Keep in mind that to exploit this, you opponent must
take the risk of playing something dubious.

>>>
>>>I don't think a full-time professional chess player is needed to do that job and
>>>actually I am not even sure if he is the most qualified to do it, unless he gets
>>>payed enough to take it very seriously :) . Judging from published comments of
>>>human masters many don't really understand strength and weaknesses of
>>>computerplayers or only in a very superficial way, although most use them
>>>regularly of course. I am convinced that people like you or Larry Kaufman would
>>>be better than professional chess players.
>>>
>>>Whether this is the optimal way I can only guess - it's what I would probably
>>>do.
>>
>>Well, it's hard to argue too much here :) but really I must. You don't want a
>>book author who has to figure everything out from scatch - you want somebody who
>>already knows it. Of course there is a wide range - from a total beginner on one
>>end, to Kasparov or Anand on the other end - but there is a huge jump up when
>>you move to someone who is a chess player by profession.
>>
>
>No doubt!! Only it is not easy to get a professional to put his heart into this
>kind of thingy unless you are wealthy enough.
>
>Using an example of my own experiments ( I really don't believe it is any good,
>but book authors are so secretive that there is nearly no common data or
>opinions availlable beyond bare scoresheets), Crafty-IsiChess was the most
>horrible experience. This was a line I had analysed quite thoroughly, but had
>just misjudged. A line that looked promising in preparation led to what I felt
>was a dead draw over the board, nearly by force. This could have been avoided by
>a higher playing level of the book cook.
>
>On the other hand, if you hire an active grandmaster you'll potentially get some
>amazing stuff, but is this really a realistic option for anyone but the top two?

Generally I don't think paying a bit is out of line. If I put a few thousand
hours of time into this, what's the big deal with a few thousand euros? I've
already got 5 Amd-64 test computers, and to the amusement of my friends my
electricity bill is 100 Euro/month :)

>
>And maybe the other side of the qualification is easy to underestimate, the
>understanding what is good for a given program and what not.
>
>Also there is so much good literature availlable that can be used that no one
>will really have to start from scratch. But yes, chessunderstanding is a limit -
>I doubt a 1400-1700 player can really do too well (confirmed also by some
>correspondence experiments done). And if you switch from an ordinary earthling
>to a 2400-2500 player there will be a compairable jump. To get a bit off-topic:
>I once played a tournament game with the black pieces against an international
>master using one of my pet lines. At some point in the game I felt I had done a
>good job and managed to equalize. In the post-mortem he pointed out that I was
>still dead lost positionally - and he was right, his reasoning was just beyond
>my abilities.
>

Yes, this is all totally normal stuff. Of course the best book author would be
Kasparov, but he's probably a bit too pricey and not enough motivated :) - so
it's just a question of drawing the line in the right place.

BTW - I thought your WCCC book was not bad. You got the better of it against
Shredder and Fritz. The one opening disaster was against one of the amateurs,
where Crafty's pieces all ended on the back rank and the king was just toasted
right in the opening - although Crafty won this game IIRC.

Vas

>And then there are ways to limit this effect - but as I might want to try it
>again some day, I want to be a little secretive too ;)
>
>Peter



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.