ELECTRONIC T-NOTES


CHESSBASE USA'S WEEKLY ON-LINE NEWSLETTER


FOR THE WEEK OF NOVEMBER 22, 1998


TEACHING FRITZ5

by Steve Lopez

Two weeks ago in ETN, we discussed the means by which some players find "holes" in Fritz' opening books and use them to defeat the program, or else use some known "anti-computer" lines (opening variations in which computers are known to play in a very weak manner).

I've spent some more time since then thinking about the issue and exploring some of Fritz' options. In doing so, I discovered an often-overlooked feature of Fritz5.

Open a game tree in Fritz5 (by hitting [F11]) and go to the "Book" menu. You'll see a command called "Learn from Database". This is an ignored feature of the program (I'd forgotten about it, too) that will help in these special "anti-computer" cases.

Let's say you have a friend who comes over to your house to play Fritz on your computer and frequently beats it by playing the same anti-computer opening traps (such as the one we looked at two weeks ago). For the sake of our example, we'll say that the friend is playing the White pieces in all of these games.

All you need to do is create a database of these games in which your friend (as White) has beaten Fritz by using these anti-computer techniques. In the main screen of Fritz, open the game tree [F11], click on "Learn from database", select the database you've created of Fritz' losses, and you'll then see a window that looks like this:


This is the window where you set the parameters for Fritz to learn from the database. Leave the "Games" settings alone; this defaults to a span of games covering the entire database. Uncheck the "Weigh player" box. Then make sure the "Black" box is checked; Fritz is Black in all the games of the database and you want Fritz to learn from these losses. Then click "OK".

The end result of this is that Fritz will avoid the variations it finds in the database in which it has been previously beaten.

There are other ways to get Fritz to do the same thing with other databases. You could create a database of games of nothing but Fritz' losses (regardless of color). Make sure the "losses" button is checked; Fritz will then learn from its prior losses and avoid those variations.

If you're training to play a certain opening, create a database of games in which that opening was played. Merge the games from this database into a opening tree. Open the tree, click on "Learn from database", select the database on your new opening, and then check the "Losses" button. Fritz will then avoid variations that lead to losing lines. To really strengthen the program in the opening, choose "Losses" and whatever color Fritz will be playing as you train against it. Fritz will then tend to avoid weak lines that often lead to losing positions. You won't win many games against Fritz this way, but you'll end up being the one trained as you learn all the nastiness that might befall you in your chosen opening (and, hopefully, learn to avoid these problem areas).

Another technique (which is especially good with large databases) is to select "Wins"; Fritz will then stick to winning openings and lines (which is quite different from avoiding losses -- in the latter case, Fritz may play drawing lines in order to avoid a loss).

Play around with this feature a bit and you'll pick up on the various techniques for using it. I'm sure you'll also notice the benefits -- both to your play and Fritz'. Until next week, have fun!

You can e-mail me with your comments, suggestions, and analysis for Electronic T-Notes. Stop by the Yahoo Chess Kamikazes Club and the Chess Kamikaze Home Page.