Using Hybrid Settings

[Norm Pollock]

Here's my idea for making a hybrid setting. It is based on the empirical and generally accepted observation that white scores approximately 33 elo points better than black. Therefore playing as black should use a more passive style than when playing as white. This is done on the GM level, I believe, but aside from the opening book moves, I don't think it is done with engines.

Here is how it would be done. Instead of an author choosing the best setting for his engine based on total games, he should choose the best setting for games played as white, and choose the best setting for games played as black. He should then use the hybrid setting of best white setting with best black setting.

After all, opening books use different perspectives for white and black, so why should this differentiation stop after an engine is out of book?

Here is a numerical example of how this might work:

Suppose there are 3 best settings: A, B and C. The "total" elo rating for the engine is 2700 for all 3 settings. The author does not know which to pick so he leaves it to the user to pick the setting.

Now if the elos are computed separately for white and black over a large quantity of games, we might get something like this:

A-white =2710
B-white = 2720
C-white = 2730
A-black = 2690
B-black = 2680
C-black = 2670

Here, the average white elo is 2720, and the average black elo is 2680 which is close to the generally accepted view that the elo difference between white and black is 33. I'm using 40 here in order to illustrate the idea with an easier to use number.

If the author were to now make a hybrid setting by using setting C for playing white (2730) and using setting A for playing black (2690), this hybrid setting would have an elo rating of 2710 (the average of 2690 and 2730), 10 points higher than each of the other settings. A small increase, but nevertheless an increase.

[ml]

Sounds like an interesting idea, and it would probably work if the engines were fairly evenly matched. If there is a significant rating difference between two engines, then the more aggressive or active settings would generally favour the stronger engine even if playing as Black. We'll need some actual game data to verify all of this.

[Mike S.]

(...) playing as black should use a more passive style than when playing as white. (...) After all, opening books use different perspectives for white and black, so why should this differentiation stop after an engine is out of book?

Actually, if the book software is good, it should not be limited to a simple white or black perspective but should recognise theory positions with switched colors.

An example (which is not realistic because it's too simplified, but I think it illustrates the logic I mean):

1.e3 e5 2.e4. Suddenly, White and Black have "virtually" changed sides.

I entered these move into Fritz in analysis mode, and could produce this variation from one of my books which does NOT contain 1.e3: 1. e3 e5 2. e4 Nf6 3. Nc3 Bb4 4. a3 Ba5 5. Nf3 O-O 6. Be2 Re8 7. b4 Bb6 8. d3 c6 9. O-O h6 etc., which is a Ruy Lopez with reversed colors.

I think the colors are not a reliable source of information, to adjust settings depending on them. For example, against me an engine can always play like white

But maybe the proposal is debateable depending on the evaluation, if it's positive or negative from the engine's viewpoint, for a couple of moves. I am not sure. Also, I can imagine that some engines already make some adjustments internally, depending on that. I wouldn't be surprised.

[ml]

I entered these move into Fritz in analysis mode, and could produce this variation from one of my books which does NOT contain 1.e3: 1. e3 e5 2. e4 Nf6 3. Nc3 Bb4 4. a3 Ba5 5. Nf3 O-O 6. Be2 Re8 7. b4 Bb6 8. d3 c6 9. O-O h6 etc., which is a Ruy Lopez with reversed colors.

The program could just set a flag in case of a transposition as in your example. With the flag set, it would simply interchange the Black and White settings, and the game would continue from there.