Forums

Fritz 12 - custom opening book training

Sort:
Elendil1357

Yep, I was on the latest GUI version of Fritz 12, and actually just got Deep Fritz 14, so I upgraded that to the latest version, and tried recreating my book, with the same effect.  However, I did discover something interesting.

First, I'm not getting my imported games from a pgn file, but rather from another chessbase db.  So when I imported the games from my repertoire db, I get the results you and I have been discussing for weeks.  For experimentation, I tried creating a new book and imported a couple games from an entirely different db, and that one seemed to work (I got [%] numbers, etc.).  So then I took it a step further and created another new book, and imported games from my repertoire db, but only a couple games in the middle rather than all 16, and it seemed to work.  So it seems as though 1 or a few games in the source repertoire db is making this break.  No idea which game it is yet.  Maybe I'll try important one at a time and see which breaks it....

Elendil1357

Ok, so strangely enough, I just tried something that worked perfectly, that was based on an idea you gave me when you said that creating a book from a french pgn file worked for you.  So instead of importing the list of games from my Chessbase database, I instead EXPORTED all the games from that database to a pgn file, then imported the pgn file into the book.  For whatever unknown reason, that worked perfectly.  That's messed up.  But at least it worked!  Thanks again for all your help on this.

EscherehcsE

Glad you got it working!

rooster85
Elendil1357 wrote:

Yep, I was on the latest GUI version of Fritz 12, and actually just got Deep Fritz 14, so I upgraded that to the latest version, and tried recreating my book, with the same effect.  However, I did discover something interesting.

First, I'm not getting my imported games from a pgn file, but rather from another chessbase db.  So when I imported the games from my repertoire db, I get the results you and I have been discussing for weeks.  For experimentation, I tried creating a new book and imported a couple games from an entirely different db, and that one seemed to work (I got [%] numbers, etc.).  So then I took it a step further and created another new book, and imported games from my repertoire db, but only a couple games in the middle rather than all 16, and it seemed to work.  So it seems as though 1 or a few games in the source repertoire db is making this break.  No idea which game it is yet.  Maybe I'll try important one at a time and see which breaks it....

I fumbled with this "feature" of ctg books for two hours now, and finally have found out what causes the issue - the engine will play a particular move based on a number in the % column (the one on the far right in book display in Fritz - btw, last three columns are not visible in Chessbase, which is what I use for book creation). When this value is zero, the engine won't play the move, ever - and that is the problem that you (and I) encountered.

Now, the % value is basically the "Prob" value (to the left of "%"), multiplied by the "weight" - the column to the left of Prob (named differently; according to the engine in use IIRC). If you haven't set any weight for a move, the Prob and % values are the same.

Now here comes the point and the crux of the issue - The Prob and % values cannot be changed manually once a book is created (% can be adjusted by setting a different weight value, however when "Prob" for a move is zero, the "%" is also zero - no matter the weight).

Thus I tried to find out, what affects the Prob value - based on the help file of my Fritz, the Prob is "the purely statistical probability of the move, based on the number of times it was played and the results achieved in the games that went into the book."

This is all good, for creating engine books - probability of choosing the moves will depend on results the move achieved... However, for our purposes (I also want to practice my repertoire against Fritz in the way you described), it's not good, because when a move has Prob = 0, engine will never play it (even though it's imported in the book).

Thus, I found the solution - it is to remove game results from games or analysis that you plan to import in the book, before importing them. I've tried importing multiple games that ended as a "Line", or even +- or =... basically anything other than the usual 1:0, 1/2:1/2 or 0:1.

When the games/analysis are imported this way, what it does it assings the Prob value evenly to all moves (as it doesn't have the results to calculate probability from) - e.g. when you have a position that has 4 moves  in the book, all moves have Prob value set to 25.0 - and engine will hence play each move 25% of the time.

 

Voila, the opening practice can begin :-)

 

Now, even better IMHO would be (aside from being able to change the damn Prob value manually after a book is created) to be able to tweak the games results in a way that would put higher Prob value to the main lines, or lines you encounter/want to practice the most (resulting in Fritz playing these lines more often than others).... However I'm still confused how exactly it assignes the value based on game result - basically, when you have a game which was lost, it assigns zero value to the first move of the losing side (and 100 to the other side), but from what I tried, I couldn't figure out easily how to tweak games results to get a non-zero Prob value for all moves (as I want Fritz to not exclude any move that's in the book)....

So for now, I think just removing games results before importing is easier and at least let's me practice my repertoire, which helps :-)

Edit - It just occured to me; after you have a non-zero Prob value (which all moves have when importing without any games results), you can tweak the "%" value by adjusting move weight - and thus changing the actual probability by which the engine will choose the move (as it selects a move based on the "%" value itself, not based on Prob value). I also think that when you have the engine book learning turned on, it changes move weight over time as well (based on results of the games played with the book).

 

Now, if only there would be an easy way to manipulate results of multiple games in Chessbase, and not having to change/remove results one by one... any suggestions on how to do that are more than welcome :-)

 

Hope my looooooong post helped

rooster85

and btw, for me this worked in the same way for both cbh and pgn databases, so no need to convert between formats....

rooster85

Edit 2 - I found the solution to my own question :-)

The easy way to quickly prepare many games for book import is to merge all games into one, save it as a new game (clearing the result first), and then to import only this one game - don't forget to have the "include variations" ticked, so it doesn't import only the main game :-)

result:

 

TheaRtOFLIONS

Hoi, ik ben nieuw op deze site dus geen idee hoe ik een nieuwe topic kan aanmaken en met de helpfunctie kom ik er ook niet uit :/

Mijn vraag : Fritz 12 / Training / Partij met rating. Ergens onderweg heb ik blijkbaar 'use openingsbook' gebruikt. Fritz zet nu sneller en komt uit op sterkere stellingen dan zonder het openingsbook. Iemand enig idee hoe ik Fritz kan laten schaken in Partijen met rating, maar dan zonder openingsbook ? Via F4 (boek instellingen) merk ik geen verschil. Thx !!

manquaman

I have Fitz 17 and this problem still happening. What a pain.