Lonely queen

[tpetzke]

Hi,

I did another test now with a Q vs NNB. The queen side seem to still have an disadvantage of 0,38 pawns.

In my previous test where the Q side was down 0,82 pawns some (not all) of the starting positions included the bishop pair. So probably in a pure bishop pair setup the value would be even a bit higher.

So the net bonus for the 3 minors (in a position with rooks present) is probably around those 0,38 pawns. The difference to the 0,82+ is very close to the bonus I assign the bishop pair already.

All tests included the rooks in the starting position (and castling rights).

Add: Just for curiosity I also gave the side with the 3 minors an extra pawn. The queen side only scored 23.8% in that case. Interestingly enough the bonus for the imbalance still ended up to be 0,37 pawns. So this seems pretty consistent.

Thomas...

[hgm]

How much bonus the imbalance needs of course depends on the base values you assign to the pieces. With Kaufman values (Q = 950) Q-NNB already is +0.25 for the minors (3x 325), which is already very close to the +0.38 you find. (Perhaps to within the statistical noise?)

In general the method always gave very consistent results for me (for opening values), as long as I refrain from deleting multiple Pawns. Not only for orthodox pieces, but also for all fairy-pieces I have tried. The reults also seem to be quite insensitive for the quality of play, as long as the engine implements the basics. (E.g. it is hard to measure the B-pair value if the engine would randomly squander the advantage of BB vs BN by trading B for B.) I used different TC, (from 40/1 to 40/10), and they all produced the same results. Which is good, as it means you can use fast TC.

To my surprise it was actually quite insensitive to the engine's idea of the piece values. No matter whether you set the Chancellor (R+N) value to Q+0.5, when you play a Q vs C imbalance the Q still comes out on top by about 0.5 Pawn. As long as both players share the misconception, this seems to cancel.

[Lyudmil Tsvetkov]

Hi,

I did another test now with a Q vs NNB. The queen side seem to still have an disadvantage of 0,38 pawns.

In my previous test where the Q side was down 0,82 pawns some (not all) of the starting positions included the bishop pair. So probably in a pure bishop pair setup the value would be even a bit higher.

So the net bonus for the 3 minors (in a position with rooks present) is probably around those 0,38 pawns. The difference to the 0,82+ is very close to the bonus I assign the bishop pair already.

All tests included the rooks in the starting position (and castling rights).

Add: Just for curiosity I also gave the side with the 3 minors an extra pawn. The queen side only scored 23.8% in that case. Interestingly enough the bonus for the imbalance still ended up to be 0,37 pawns. So this seems pretty consistent.

Thomas...

Thanks.

How do you ensure randomness of the ICE moves? Is ICE playing random?

You measured, from what I understand, 40cps advantage for the 3 minors in a position with additional pieces. My impression has always been that the queen has an advantage in most cases of the imbalance with only queen and 3 minors present (excluding 2 bishops), but probably an additional piece already gives additional edge for the bigger number of pieces. It would be interesting to repeat the test with a stronger engine. (of course, I do not suggest that you do this)

Relying on your data until now (and I hope new data with rooks involved in the imbalance will shed new light), we have:

- bonus for the 3 pieces in case of a Q vs 3 pieces imbalance - 40cps
- standard 2 bishops bonus remains the same at 50 cps
- if able to measure the difference in performance in a pure Q vs 3 minors case and in the case of Q+R vs 3 minors + R, we will even probably know how much value an additional piece adds to the score of the more numerous pieces (I suppose it might be some 10-20cps)

Now we will all be waiting eagerly to know from you, Thomas, how 3 pieces perform vs the queen, when there is one rook, or 2 rooks among those pieces.

Many thanks again for testing and sharing. I hope your findings might be useful also for other people.

[tpetzke]

How do you ensure randomness of the ICE moves? Is ICE playing random?

You mean like a Monte Carlo simulation ? No, I don't do that. I'm using my latest development version. As I want to use the results I'm using my engine for the test.

I don't think that a different engine will produce different results anyway, maybe the quality of the played games increases but the W-D-L numbers should look very similar.

To ensure that not always the same game is played over and over again, I have created a bunch of different starting positions with the same imbalance. In addition when played at fast time control the computer background processes introduce some diversity.

Thomas...

[Lyudmil Tsvetkov]

I don't think that a different engine will produce different results anyway, maybe the quality of the played games increases but the W-D-L numbers should look very similar.
Thomas...

That is questionable, because we had a case when Texel and Queeny produced very different results.

In any case, what you have started doing is simply marvelous, I very much hope you will share all the results.

[tpetzke]

HGM states:
Queeny uses a value 500 for the Knight in stead of the regular 325 (where Q=950, R=500, B=325 etc.), and I am pretty sure that this will totally wreck it for playing normal Chess, as it will seek to trade Rooks for Knights, and Queens for two Knights.

I mean chess engines designed for regular chess will produce similar results if they have a decent evaluation.

Thomas...

[Evert]

just for the record I did the measurement test and documented it here

http://macechess.blogspot.de/2013/12/qu ... inors.html

Seems your 0,70 pawns guess was actually pretty good.

Very interesting indeed!

In Jazz, I use the bad-trade idea from Crafty (implemented in the same way), which awards a penalty of 50 cp to the queen-vs-three-minors imbalance (for the queen side; Crafty has a 42 cp penalty for this case). Taking into account the raw piece values (but not the bishop pair bonus) this gives an overall penalty of -75cp for the queen side.

This is only very weakly tuned, however; perhaps I should try measuring the imbalance in a similar way and see what comes out of it. From the sound of things, the current guess may not be too bad.

[tpetzke]

Hi,

I'm done with my tests of queen vs 3 minors. I performed several tests where I altered the number of rooks on the board.

Here are the numbers

http://macechess.blogspot.de/2013/12/qu ... art-2.html

[hgm]

Wow, this seems like a really spectacular demonstration of non-additivity. One should be careful not to immediately accept the score percentages without further checking, however: as the number of pieces thins, and the number of Pawns stays at 8, it becomes more and more difficult to keep all the Pawns protected. A Queen can exploit that much faster than the minors, as it can start forking Pawns in just two moves. So there could be some worry whether all positions used in the test are actually 'tactically quiet'. If there is a fraction where the Queen side immediately gains a Pawn, this would obviously bias the results in an undesired way. In 'typical' Chess positions where this imbalance could occur, the Pawns would never form a closed rank, but rather chains where many of the Pawns would be protected by other Pawns (and there very likely would not be 8 on each side...).

Anyway, the elephantiasis effects predicts this trend: Rooks for the minors side hinder the Queen, and have little to fear themselves in terms of LxH attacks, while Rooks of the Queen side are a liability, being constantly harassed by the minors. With two Rooks each this effect should double, but with 2-vs-2 the opportunities for Rook trade are expected to be 4 times more frequent that with 1-vs-1, and this probably ameliorates the effect. Playing an RxR-avoiding strategy by the minors side would devaluate his Rooks more than it is worth.

[tpetzke]

A Queen can exploit that much faster than the minors, as it can start forking Pawns in just two moves. So there could be some worry whether all positions used in the test are actually 'tactically quiet'.

Yes, I cannot rule that out completely. As the queen is behind the pawns it usually requires White 2 moves to bring the queen into play, in most cases enough for Black to defend against the immediate threats.

If I build pawn chains and stuff it does not necessarily become easier to rule out tactics but now other stuff like rank boni, central control, pawn protection, weak squares start playing a role too, so the value of the imbalance gets polluted.

Thomas...