Customizing the Openings/Endgames Training Databases in ChessBase’s Rybka Interface: Preliminary Results and Tips
I have just started tweaking the opening and endgame training databases that shipped with my ChessBase version of Rybka 3. Here’s what I’ve determined so far:
1. Work with a BACKUP of the database. I have never had a ChessBase error so serious that it corrupted a database beyond repair…until just now. I’ll have to load my Openings database off the CD-ROM again.
2. Modifying the openings database seems easy enough. The computer recognizes an opening variation by the “White Player” field, and a sub-variation by the “Black” player field. For example, here’s the Sicilian section of the database:
The column headings were cut off in this screenshot, but “Sicilian Defence” is the “White Player” field, and the column to the right with the variation names is the “Black Player” field.
When practicing the opening, you are given the option of a general or a detailed list of openings. The default general openings list is generated from just the “White Player” field, so we have a generic Sicilian with just 1.e4 c5 and no “Black Player” entry. Under that, we have sub-variations of the Sicilian with the “Black Player” field giving the variation’s name.
If you add new openings to the list, they will automatically be alphabetized within the engine interface. For example, I added a “Colle-Zukertort” as a test case to the end of the database, and in Rybka it was automatically put in the openings list under the Cs.
You can include sub-variations, and influence how often the computer randomly chooses that line. For my Colle test-case, I entered three lines:
Using a “!” annotation doubles the likelihood that the computer will select that line. A “!!” triples it. Other annotations such as “?” or “?"!” mean the computer will NOT pick that line (so not that useful, unless you want to temporarily turn a line off).
In the above case, the computer would pick the “Colle” (1…d5) order twice as often as the “Zukertort” (1…Nf6) move order. Of the two “Colle” lines, the computer would pick either 2…Nf6 or 2…e6 3.e3 c5 randomly. Whichever line the computer chooses, it will play forwards to the end of the line, at which point you get to play against the computer.
The endings database was much harder to figure out. It was not as simple as substituting my “100 Endgames You Must Know” database. It took me a long time to figure out why Rybka was saying that my database wasn’t valid: I tried stripping annotations, stripping variations, truncating the games, and finally discovered:
The position cannot have an evaluation. No “1-0”, “+-” etc. Just “Line”.
Since I routinely include these evaluations (just as endgame books routinely indicate whether white to play wins, loses or draws), this means that the database will require each game to be re-saved without the evaluation.
As with opening training, the computer zips to the end of a line before play begins. In a lot of endgames there is one “best” move (only drawing move, or move offering greatest resistance), and I routinely record all the lines and their evaluations. I was hoping the computer would follow a line, or select a variation, and play it against me stepwise. Unfortunately, it seems the best you can do is save the bare positions and play them against the computer.
I’m going to just continue entering the contents of 100 Endgames You Must Know into my database as usual, and at some future point I’ll consider converting it to a Rybka-friendly database.