Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Opening Books / Sick Panda.ctg
- - By Auryn (***) [it] Date 2010-03-27 22:55
Hello Guys :-)

here is the link to the book I made for playing in the sedat tour, sick panda.ctg. --->>

Ok, let me give some information about it:

First of all, I have to confess that I am a very unexperienced imported games book maker, as I have always only worked on my all-manual book, sikanda. (that's why the name sick-panda, as I tried to create a clone of my book called sikanda)

Because of this inexperience of mine, I thought I had to try something different if I wanted to create a competitive book, so an interesting idea came to my mind: why not create a book based only on my own games? I thought this could have been interesting and fun, and most importantly this would have saved me a lot of filtering work, needed thing to do (I think) when you import all kind of games played by who knows who. So I imported almost 10.000 of the games I played on the Playchess server; every game has been played by my nicks there, I made some very little filtering, the most important one was to only import games played since the beginning of year 2009 until march 13th 2010, date when I started creating the book. Every time control game has been imported, from 3 minutes to 180 minutes and even longer, played by all the sort of engines I have used, and 99% of those games have been played on slow computers.
While at first sight it may seem dubious to trust games played on slow computers, on the contrary, I believe that those games are more reliable than games played on fast machines, simply because a win in those conditions is most of the times a "true win", not a "power win". It indicates that a good opening had been played, otherwise a slow pc often with a weak engine wouldn't have been able to win. Oh, I almost forgot: I only imported my wins when I had white, and my wins and draws when I had black.
After doing this, I manually selected in green the main lines (especially the first moves) I play with both colors, in order to be sure they are played.

Ok, I hope you find this book at least interesting. It has though some weak points:

1) Don't use it if you are not an absolute "all-the-time" Najdorf player. This book only and always plays that opening with both colors.
2) Same with other main openings, my main book (sikanda) plays very few openings, and as sick panda is a clone, you will have to accept to always play those few openings, like them or not :-)
3) It's based on less than 10.000 games; this means that if your opponent doesn't play main lines you will probably be out of book very soon.
4) forget the idea of trying to win as black, this book's main aim is to easily draw as Black, some variations in Najdorf go straight from opening to draw tablebase endgames. As White though it tries to win; that's why I made an exception for the white side after 1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 a6: now my main move as white should be 6.Be3, but as this would always only lead to a tbs draw, I decided to select in green the move 6.h3, as it's still an unexplored variation and can give an engine as white the chance to find an advantage.
Nothing else comes to my mind at the moment, maybe Nelson can help? :-)

Ok, it has also some strong points:

1) it's very useful if you want to prepare against me for a future freestyle event ;-)
2) I tried some games on playchess, and it seems not so bad, I got 9th place with a 40 kns rybka in a prize tour today (march 27th) (I beat alan, tsk tsk ;-)
3) as also many long games are imported, it should play decently in long games as well, not only in blitz.

The suggested settings (but I am not sure if they are the best) are: 

All the three bars all to the left.
Minimum games: 0
Moves Limit: 100

Parent - By BrandonSi (***) [us] Date 2010-03-27 23:45
Wow, thanks for taking the time to tell us about Sick Panda. :)

Best of luck in the tournament!!
Parent - By Sebastian Boehme (**) [de] Date 2010-03-27 23:48
Hi Eros,

nice to finally see one of your books being published to the mass audience.

I look forward to refuting its lines in my book cooking studio. :-O

Parent - - By Nelson Hernandez (Gold) [us] Date 2010-03-27 23:50
I have nothing to offer but blood, sweat, toil and tears, Eros.

Assuming that books now entered are final and nobody is going to sneak in new ones...I elected to play Larsen with white (again), but entirely different lines this time with considerable variety in the tree.  It should play pretty well.  With black I will be playing a combination of Pirc and King's Indian with the ability to handle unusual sidelines.  The idea is to score about 50% with black and 55-60% with white.  An overall score in the 53-54% range is successful for a .ctg book against clever opposition.  Games used: all kinds.  Computer, human, you name it.  The games aren't as important as what deviations from main lines to choose, and these I chose pretty carefully.  There are definitely still some holes in there and I expect to lose a fair number but I think the smaller number of games played between books helps me as I will be less neutralized by book-learning.

1.b3 is a really good opening for white for all the reasons that Alan could tick off.  You take control of the game and, if you specialize in this opening, you will have a much deeper book than your opponents.  The problem with it is that if you become too predictable you become a fixed target.  That's my main worry this time: that someone will have come up with some well-researched, unusual sidelines to throw me off.  Undoubtedly one or two will have done that, and that is why next time I will most certainly be ready to play against 1.b3 with black pieces and on to something else with white.

The Pirc/KID theme is not particularly strong for black and not naturally successful without a lot of help.  When I first started testing the unedited book was getting slaughtered by its sparring partners and things looked quite discouraging.  But systematically I plugged the recurring losing moves I was making and by the end it was holding its own against more recent books, exiting book with rough parity and a clock edge, consistently forcing opponents to play long games. 

Once again I don't expect to win as most people will still play B90 and get massacred by Eros and Adriean, giving both of them a disproportionate number of easy wins.  Whoever is the biggest B90 stud will still probably win this tournament because there are still too few deviationists.  But gradually, if the number of dissidents and free thinkers grows, and they prove themselves in combat, the B90 stranglehold may finally be broken among the most influential bookmakers and we'll enter a new era.  That's what I hope for, anyway. 

Good luck to all.
Parent - By BrandonSi (***) [us] Date 2010-03-28 00:01
Interesting Nelson.. I think we may end up with some great games between you and I.. We have extremely similar strategies.
Parent - - By Banned for Life (Gold) Date 2010-03-28 00:56
OK, now I must rationalize losing against Eros and his 40 kn/s museum piece...

Truth is, I've been using only my own games for a while now, but I use them all, win, lose, or draw. This would be OK, if I had some kind of clue as to how to control a ctg book, but after seven years, this is still beyond me. So my game with Eros, rather than following a path into B35 where I think I could have prevailed, went into a B77 line that Eros has known how to crack for several years now. The problem is that although Eros wins easily against this line by booking cross castling and aggressively sending the g and h pawns, most others still lose to it, so it still has a high score and tends to get picked even if I set its initial move weight to -125. I've also had issues with the book not playing certain moves, even though they have been played hundreds of times with good results.

So I've come to the conclusion that the ctg book format is really shit, but even so, losing to Eros is totally unacceptable. So in the future, I guess I will have to eliminate moves that lead to this variation, or play with tournament book checked (which seems to make it more likely that the book will ignore moves that are not green).
Parent - - By BrandonSi (***) [us] Date 2010-03-28 02:00
[play with tournament book checked (which seems to make it more likely that the book will ignore moves that are not green). ]

That's actually what it does.. It uses green moves, doesn't use red moves.. If you have two green/red, it uses weighting.

I understand your frustration with the engine not playing the moves it should be, according to the book.. I've seen this a lot myself, drives me crazy.
Parent - By Banned for Life (Gold) Date 2010-03-28 05:26
I'm referring to cases where there are a bunch of moves that have been played frequently and successfully, but none is colored green. When the book is first compiled, a move will be picked based on number of games, success rate, and book settings. The distributions are shown in the two percentage columns. After using the book for a while, many positions will have all zeros in the percentage columns and won't play a move unless it is green. I guess the book database is damaged at this point.
Parent - - By Auryn (***) [it] Date 2010-03-28 02:06
I am also starting to learn how bad the .ctg format with imported games is: I just realized that if a move is colored in black and it has like less than 30 imported games, if that move doesn't win in a game, the next time it won't be played anymore! The book will just ignore that book move, even if the statistics are all wins! That would be a real disaster for a book like mine, and I predict already many losses in the sedat tour because of this... Anyway I think there is a solution: for those who downloaded the sick panda, try to open the folder containing the 4 files, go to property, and mark "read only". I found out in the past that that should avoid all kind of learning (or should I call it "mislearning"?) and so it should avoid that problem I was talking about of the ignored moves with less than like 30 games if they don't win the first time.

Parent - By Nelson Hernandez (Gold) [us] Date 2010-03-28 03:55
Anson and I agreed that .ctg was horrific and totally unusable in serious competition as far back as 2004.  What is really desperately needed is a whole new book format that takes the best of what is already out there and combines it with some more progressive ideas.  Most of all that format has to be able to handle tens of millions of database games without crashing.  I am confident somebody will invent this someday for public use.  I consider it a wide open field for development.
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-03-28 03:51
Moves I want to play I mark green and add !! 
Moves I don't want I mark red and add ??  I do this for any move that has >0% in the far right column. 

It is tedious to do this for every position you're likely to see, and the technique is probably overkill.  But I haven't had any move-selection problems with this approach. 

I don't eliminate book learning.  It helps me see what has worked lately in the far reaches of the tree.  If you don't like the book learning you need to neutralize it with the double-barreled approach.
Parent - - By Banned for Life (Gold) Date 2010-03-28 05:23
Most of the moves marked !! in my book are marked red and represent discoveries that I wanted to save for important games. Of course, not being very organized, I have no way to get back to these positions...

Normally, I add in games after collecting about 50 new ones. After doing this a few times, I start getting positions where both percentage columns are all zeros no matter how many times the position was played from successfully. Usually a green move will still be played, but moves that are not green will be ignored (i.e. engine will select the move). Maybe this is related to the 30 game concept that Eros mentioned. I'll try making the book read-only during play to see if that helps. Of course that isn't consistent with book learning, but having to color every desired move green takes too much time, as we probably all know from experience.
Parent - By Nelson Hernandez (Gold) [us] Date 2010-03-28 11:47
Booking takes too much time, period.  But if you want it to work, you waste it joyfully.
- By ppipper (*****) [es] Date 2010-03-30 18:34
managing books with Aquarium has fresh ideas and a new totally different approach and IMO pretty advanced.

Editing and tunning is much more powerful than doing so with ctg formats. It is not my case, but if you see learning features as an advantage, Aquarium does not support them.

I wonder how could I participate in one of sedat's tournements to test these ideas but I dont find how to without delivering all the trees
Up Topic The Rybka Lounge / Opening Books / Sick Panda.ctg

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill