FOR THE WEEK OF June 4, 2000


by Steve Lopez

Most of the available engines for use with the Fritz interface (and the other 32-bit interfaces) have configirable engine parameters, in which you can tweak the engine's "thought processes" to alter its style of play. By far the most tweakable of these engines is Nimzo7.32. Programmer Chrilly Donninger has provided us with a lot of ways to modify Nimzo's playing style.

One of the most interesting is the ability to revalue the pieces. In fact, looking at the default engine parameters in Nimzo7.32 can tell us a lot about the program. Upon loading the engine and opening the parameters window, we see a series of boxes with numerical values for the pieces:

Right away we notice a couple of interesting things. Nimzo uses a slightly different scale for its internal evaluations than what most computer chess players are used to. We usually see the pawn valued at 100. Nimzo gives it the slightly higher valuation of 110. We also see that Nimzo values Bishops just a hair higher than Knights (rather than the half-pawn or quarter-pawn value given in most chess books).

A closer examination reveals further interesting data. The conventional wisdom in chess is that a Knight is worth about three pawns and a Bishop is up to a half-pawn more valuable. But look closely at Nimzo's valuations. With a pawn valued at 110, you would think that a Knight would be worth 330 and a Bishop worth up to 385 (depending on what value you place on the Bishop). But Nimzo values its minor pieces more highly: 420 for a Knight (almost a full four pawns) and a Bishop is only a hair more valuable than a Knight (425). What does this tell us about Nimzo? First, you'll likely see Bishop for Knight trades occur fairly frequently, particularly when such trades disrupt the opponent's pawn structure. This provides a bit of positional finesse for the program in some cases (the program is capable of creating pawn targets). On the other hand, this may backfire in the case of minor piece trades in the opening, before the position is fully developed and the true relative values of the Bishops and Knights become known. Second, you'll see cases in which Nimzo attempts to preserve a minor piece for use in the endgame. While many programs would be content with a trade of a minor piece for three pawns, Nimzo is going to want four pawns in compensation for the piece; you may see this as a trend of the course of many games with Nimzo.

Most chess players will tell you that a Rook is worth five pawns. In this case, you'd expect to see a Rook valued at 550; Nimzo, however, values a Rook at 685, which means that a Rook is worth more than five pawns in Nimzo's way of thinking -- in fact, Nimzo thinks that a Rook is worth more than six pawns. Again, you'll see this effect the program's game on a couple of levels, not only in cases of possible Rook sacrifices, but in cases of Rook vs. minor pieces and pawns. Most programs will consider giving up a Rook for a minor piece and a pawn, but Nimzo thinks that it's not enough compensation (Bishop plus pawn is equal to 535 compared to 685 for a Rook -- in fact, a Rook is more valuable to Nimzo than a Bishop and two pawns [645]).

Then we get to the Queen. Nine pawns is the conventional value for a Queen, but Nimzo thinks the lady is more valuable than that: 1240 versus the expected 990. In fact, a Queen is worth more than eleven pawns to Nimzo. Note, though, that a pair of Rooks is still more valuable than a Queen (which follows conventional endgame wisdom -- in fact, two Rooks is still more valuable than a Queen and pawn to the program).

You'd have to ask Dr. Donninger why these values were selected (I don't have that info), but these were evidently good choices. Nimzo7.32 scores very well in competitions with other chess programs.

What's important to us as users of Nimzo is that we can play around with these values and alter the program's playing style. For example, if you find Nimzo making Bishop for Knight exchanges that you view as unsatisfactory, you can revalue the Bishop to a higher number. Of, if you'd like to experiment with the "Queen vs. two Rooks" debate, you could increase the Queen's value to 1370 to see what happens.

Another experiment to try is to revalue the pieces along more conventional lines: pawns at 100, Knights at 300, Bishops at 325 to 350, Rooks to 500, and Queens to 900 and see what effect it has on Nimzo's play. Or if you'd like Nimzo to prefer having the Bishop pair in the endgame, you can crank up the value of the Bishop so that Nimzo will tend to hang on to them (and you'll also notice a "Bonus Bishop pair" value that you can also experiment with, just to see what happens). In fact, this may have the effect of making Nimzo a "Bishop-hunter", in which the program tries to preserve its own Bishops while capturing yours.

To change a piece's value, just left-click in the proper box and change the number listed there (either by typing it in or by using the arrow buttons displayed in the box). Make sure you click "Save" to put the changes into effect. You can always return Nimzo to its default settings by clicking the "Defaults" button. Note, too, that engine parameters aren't used in ChessBase 7 -- a program analyzing in ChessBase always uses the default values.

My personal position on all of this is that I'm not a big fan of revaluing pieces, other than as an experiment or just to play around and have fun with it. I once had a user ask if it was possible to revalue pieces for studies of certain endgames (for example a bad Bishop plus pawns versus a good Knight plus pawns, in which you would decrease the Bishop's value). I've found that many modern chess programs already take these changes into account as part of the standard algorithm, so user-initiated changes really aren't necessary (and may skew the results when one is performing "serious" analysis). But, as always, your mileage may vary and you might even discover a range of settings that help Nimzo's play a great deal. Remember, it's your program -- so tweak to your heart's content. That's why the settings are there.

Until next week, have fun!

You can e-mail me with your comments, suggestions, and analysis for Electronic T-Notes. If you love gambits and sacrificial play, stop by my Chess Kamikaze Home Page and the Yahoo Chess Kamikazes Club.