Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old January 17th 04, 07:10 PM
Derek Wildstar
 
Posts: n/a
Default Engine tournament details:

When attempting a engine tournament (Frtiz7 gui), is it wise to globally
link the opening book learning for all engines, or should each engine have
it's individual configuration, specifically book treatments. I have an
immense book, self-crafted of 450K quality games, weighted.

In this case the desired result is to see where the strengths and weaknesses
of each engine is, in particular, to each other, not necessarily in general.


  #2   Report Post  
Old January 17th 04, 08:32 PM
CeeBee
 
Posts: n/a
Default Engine tournament details:

"Derek Wildstar" wrote in rec.games.chess.computer:

When attempting a engine tournament (Frtiz7 gui), is it wise to
globally link the opening book learning for all engines, or should
each engine have it's individual configuration, specifically book
treatments. I have an immense book, self-crafted of 450K quality
games, weighted.

In this case the desired result is to see where the strengths and
weaknesses of each engine is, in particular, to each other, not
necessarily in general.




You could well argue that as most top engines the programmers work in
close cooperation with opening book creators to optimize the specific
strenghts of their engines, letting each engine play with its own book is
the most preferable way.

--
CeeBee


"I am not a crook"

  #3   Report Post  
Old January 18th 04, 12:00 AM
HD
 
Posts: n/a
Default Engine tournament details:

Derek Wildstar skrev:
When attempting a engine tournament (Frtiz7 gui), is it wise to globally
link the opening book learning for all engines, or should each engine have
it's individual configuration, specifically book treatments. I have an
immense book, self-crafted of 450K quality games, weighted.

In this case the desired result is to see where the strengths and weaknesses
of each engine is, in particular, to each other, not necessarily in general.


Even if I'm "just" checking out specific lines, I create seperate, but
identical books for each of the partipants, that, of cource means each
version number too. To get some good "products", its seems quite obvious
to me that when learn-values from one engine, easily makes another play
diffently - may it better or worse. I've tested it more than once. If
I'm to estimate how the single engine handles all kinds of positions as
a product of those lines, I can't say that with certainty without having
at least one for book each.

Playing matches and tournaments, there's doubt with me either: The
original books are best; however, with some experience its possible to
improve them, and sometimes I'm able to do that with the games from some
particular other engine(s), which is handling some significant
middlegame approaches just as well, and sometimes better.

Therefore I have (tried to) improve the books for all the versions of
all the engines I'm using. And for freeware-engines, I'm always trying
to make CB-books, as well suited for every single version - to give them
a fair start as well. It takes both a lot harddisk space, and great deal
of time to make a reasonably proper book for an almost unknown engine,
but engines can be very different from one version to another - in both
the overall strenght as well as in playingstyle: Shredder 7 and Shredder
7.04 are extreme examples.

HD



  #4   Report Post  
Old January 18th 04, 06:02 AM
Derek Wildstar
 
Posts: n/a
Default Engine tournament details:


"CeeBee" wrote in message
. 6.67...

You could well argue that as most top engines the programmers work in
close cooperation with opening book creators to optimize the specific
strenghts of their engines, letting each engine play with its own book is
the most preferable way.


Any speculation as to the level of detriment sharing a global book? I can
say this already, I've seen 3 Danish Gambits already by three different
engines, a loss for white each time (surprise.) But now, that line is
clearly in the deep red according to the learning column.

The thought of tuning several books optimally almost makes me want to not do
it and instead run several hundred "depth -1" engine tournaments, which do
little than get the engines out of book, evaluate to a shallow depth, and
them move on to the next game. Neat feature, eh? Would you agree it would
serve to chunk a few lines?

It's not just the workload this would seem to take, but I have no
confidience in my ability to tune books to each engines ideosyncracies, I
wouldn't even know where to begin, and I ususally do. Bah.

I tend to follow your reasoning Mr. CB, lay it on me.



  #5   Report Post  
Old January 18th 04, 06:07 AM
Derek Wildstar
 
Posts: n/a
Default Engine tournament details:


"HD" wrote in message news:z0kOb.75356

Therefore I have (tried to) improve the books for all the versions of
all the engines I'm using. And for freeware-engines, I'm always trying
to make CB-books, as well suited for every single version - to give them
a fair start as well. It takes both a lot harddisk space, and great deal
of time to make a reasonably proper book for an almost unknown engine,
but engines can be very different from one version to another - in both
the overall strenght as well as in playingstyle: Shredder 7 and Shredder
7.04 are extreme examples.


Good points, and in case you didn't read CB's reply, I'll ask you this same
question: Assuming I can't optimize each engine's book by hand, for reasons
real and imagined, would a massive engine tournament with a "move limit of
1" (I mistakenly called it a depth of -1 prior) serve the general same goal
of improvement by 'prelearning' bad lines...or at least, engine-evaluated,
bad lines?

If you are so inclined, could you go over your methodology of hand tuning?
Perhaps focus on specifics of Shredder 7 and 7.04, for example.





  #6   Report Post  
Old January 22nd 04, 10:22 PM
HD
 
Posts: n/a
Default Engine tournament details:

Derek Wildstar skrev:
"HD" wrote in message news:z0kOb.75356

Therefore I have (tried to) improve the books for all the versions of
all the engines I'm using. And for freeware-engines, I'm always trying
to make CB-books, as well suited for every single version - to give them
a fair start as well. It takes both a lot harddisk space, and great deal
of time to make a reasonably proper book for an almost unknown engine,
but engines can be very different from one version to another - in both
the overall strenght as well as in playingstyle: Shredder 7 and Shredder
7.04 are extreme examples.



Good points, and in case you didn't read CB's reply, I'll ask you this same
question: Assuming I can't optimize each engine's book by hand, for reasons
real and imagined, would a massive engine tournament with a "move limit of
1" (I mistakenly called it a depth of -1 prior) serve the general same goal
of improvement by 'prelearning' bad lines...or at least, engine-evaluated,
bad lines?

If you are so inclined, could you go over your methodology of hand tuning?
Perhaps focus on specifics of Shredder 7 and 7.04, for example.



No doubt it'll be massive tournament - or perhaps a match with the
engine in question!? Sure I it'll work too, whether reason for
conducting it are real or imagined. I've thought about it myself, but I
think the time is more well spend on certain lines - openings and
variations, and to avoid what seems to be bad in any case, investigate
interesting ofbeats a bit: Seek out what some particular engines prefer
to play and how they handle certain themes.
I believe it leads to some faster results, suitable for the engine - but
not necessarily more solid. One thing I'm very sure of though: It's more
fun!!

Handtuning is something I do when I have my products: The games. From
the behavior of an engine in a bunch of games, I try to estimate how
well is plays according to some chosen criteria, and thereby derive a
prober set of opening lines. This could be how well it handles the
pieces on the queenside in the late middlegame/endgame (Shredder 7.04 is
very strong at this area), and find the lines that most likely will lead
to this kind of play, and then narrow it further down. I try to forbid
lines that are unsound for the engine, and by statistics encourage it to
play the sounder lines, but avoiding force: Instead I use the games from
my tournaments and matches (and sometimes some of my own ideas!).
Creating books that only works at some later lines (not just one
position), and having some different engines work on them, brings both
the good and useful, as well as bad lines to my desk in a selective
maner. From that I can select.

Sure, whatever you do, I takes time!
If the answer isn't good enough, feel free to ask again - but I cant'
promise that I'm able to give a better one.

But I'll happy to hear what you do, and how it all turns out!

HD

Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
christmas-tournament 2003: a 15 rounds tournament on 1200 mhz and 40/120 time control Thorsten Czub rec.games.chess.computer (Computer Chess) 2 January 17th 04 01:25 AM
Tournament round generator out there for chess players? Alberich rec.games.chess.computer (Computer Chess) 2 November 21st 03 11:41 PM
validation for a new chess engine Ken Stauffer rec.games.chess.computer (Computer Chess) 1 November 15th 03 05:17 PM
Shredder6 engine help. Gian-Carlo Pascutto rec.games.chess.computer (Computer Chess) 1 September 22nd 03 04:37 PM
Fritz 7: How to let engine play a Setup Position John Cordes rec.games.chess.computer (Computer Chess) 1 August 19th 03 11:14 PM


All times are GMT +1. The time now is 01:56 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017