Author: Amir Ban
Date: 06:34:07 09/09/98
Go up one level in this thread
On September 08, 1998 at 23:36:05, Serge Desmarais wrote: >On September 08, 1998 at 19:16:58, Robert Henry Durrett wrote: > >>On September 08, 1998 at 12:24:07, Don Dailey wrote: >> >><snip> >> >>>I think it is still fundamentally superior to the other programs. It >>>may not actually be the very strongest currently, but this may be because >>>Richard has not made any substantial effort to stay ahead and I also >>>don't think the book is engineered as well as the top contenders, which >>>could be a big part of the reason his program is not dominant right >>>at the moment. >>> >>>- Don >> >>Your observation that you don't think "the book is engineered as well as the top >>contenders" sounds like an interesting concept. I get the impression that >>playing Genius with the Fritz book would not completely null out the Fritz >>advantage because, I assume, the Fritz book is "engineered" for Fritz. I >>presume that this means that the Fritz book lines are selected to optimize Fritz >>performance by covering up specific weaknesses of Fritz and capitalizing on >>specific strengths of Fritz. If this were so, then forcing Genius to use the >>Fritz book would not be optimum for Genius because Genius's specific weaknesses >>and strengths might be different from those of Fritz. >> >>If the above is an accurate portraial of what you meant by the book being >>"engineered," then forcing Fritz to play with the Genius book would be equally >>non-ideal. >> >>I wonder if a match between Genius and Fritz, with Fritz being forced to use the >>Genius book would result in more equality in the win/loss/draw results of such a >>match. In other words, if the problem were primarily one of book engineering, >>such a match would point this out. >> >>Am I anywhere near close to what you were thinking? > > >Here is how are made the Fritz books : you take a database of HUMAN games of >relatively high quality (as they say/suggest in the Fritz book) --which means >the players have a high rating and the time controls were 40/2 or the like-- and >you import these games into the/an opening tree. Of course you do not import ALL >the moves of each games, but EITHER a certain FIXED number of half-moves ofor >all the games (between 1 and 99) OR ECO-like lenght PLUS a certain number of >half-moves for all the games (between 0 and 99). > >Now, BESIDES the moves played by humans, Fritz also import the game scores, >number of games, average elo rating of the players and links these to every >respective moves. For example, you import 100,000 games into the tree ; in >60,000 of these, the first move played was 1.e4 ; 1.d4 was played 30,000 times >and so on (these number are fictitious). Let's go back to the 1.e4 move (played >60% of the time. The average rating of the players who chose that first move was >2540 and their rating performance with that move was 2580. Of all the 60,000 >games 25,000 were wins, 15,000 were draws and 20,000 were losses. > >Now here is HOW Fritz picks its moves : at base, it takes the popularity of the >move (% of times played) and then balance it with the % of wins (score), ratings >of the players and performance rating of the moves (don't ask me the >mathematical formula!) to decide a percentage of chance of playing it in a game. >Later, it will refine this percentage by adding a value based on its "learned >experience" when playing that move (you can also manually change the scores). >But you have to know that Fritz will NEVER play a move that was played only ONCE >if there are much more popular choices. It will NEVER play a move that has lost >100% of the time or has a sensible negative score (UNLESS it is the ONLY move in >book for that position, and even in that case it would have avoided the line >earlier). In the Fritz basic book, there are NO preset values (manually >adjusted) for any move. It contains over a million moves/positions with all the >known main lines for every major opening. > > >Conclusion : Except for a careful chosing of the games imported in the basic >book, there is NO manual fine tuning of Fritz book by the programmer! But the >program will/should do it automatically. Winning more often with the moves that >best fit its "style", the learning will be very positive, thus ending in more >and more chances of playing these moves, if not EXCLUSIVELY these moves! > > >I think Crafty uses a similar process and I think I remember Robert Hyatt saying > that Chessbase kind of "copied" Crafty's book learning for Fritz. > >Serge Desmarais What you describe here sounds pretty accurate, but is not what is called "book learning". "Book engineering" sounds more accurate to me. Perhaps the Fritz book learning came from Crafty, but not this. To my knowledge, no one has been as methodical in the automatic generation of books, or put it on such a sound logical basis (as you describe) as ChessBase. They should be allowed the credit for that. Amir
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.