Lonely queen

[tpetzke]

Hi,

this was only a first test to see whether the whole method produces meaningful results. It seems it does, so I will do some more stuff here.

Currently I'm doing a test of Q vs RR+Minor. With both 8 pawns the queen side only scores 14.9% (which is not surprising). I will rerun the test with giving the queen side an extra pawn. We'll see.

Thomas...

[hgm]

It is a bit unclear to me what you guys mean when speaking about 'handling the imbalance'. Normally imbalances 'handle themselves', because you just add piece values. So it seems there is an underlying assumption here that piece values are defined by fitting them to other imbalences than Q vs 3 minors, and that the thus obtained values are somehow no good in the case at hand. But of course Q vs 3 minors could have been used to define the Queen value.

Kaufman values (with Q=975 not the way he corrects it to 950) would have Q exactly equal to BNN (3x 325). Lyudmil is correct in remarking that against BBN the minors have an extra advantage of 50; an imbalances Q-BBN is worse for the Queen as Q-BNN.

In both cases the minors appear to have the advantage. I think that is also what common wisdom says. This would indeed suggest that 975 is an overestimate for Q, and 950 would be better.

Note that if 'elephantiasis' plays any role here, it is probably already incorporated in the base values.

Using this method, I always have difficulty in getting the correct value of Rooks and similar pieces (which are difficult to develop from behind a closed wall of Pawns). They always seem to test ~25 cP too low compared to what you would expect. Of course we all know that Rooks benefit from open files (perhaps as much as 25 cP?), and starting a Wazir (a 1-step Rook) in front of the Pawns indeed make it test 25 cP stronger than when you trap it behind the Pawns. Of course for a Rook starting it in front of the Pawns would not be very helpful, as it immediately gets harassed by minors, and might even be lost to them. It is just too valuable for that. (While a Wazir is only ~130 cP.)

[Lyudmil Tsvetkov]

Hi,

this was only a first test to see whether the whole method produces meaningful results. It seems it does, so I will do some more stuff here.

Currently I'm doing a test of Q vs RR+Minor. With both 8 pawns the queen side only scores 14.9% (which is not surprising). I will rerun the test with giving the queen side an extra pawn. We'll see.

Thomas...

Hi Thomas.

Here are 8 testing positions for the imbalance (I excluded the one you already did).

1. [d]rn2kbnr/pppppppp/8/8/8/8/PPPPPPPP/R2QK2R w KQkq - 0 1
I guess the queen side here has the advantage, so probably you should exclude only this one, with 2Ns + B from the imbalance.

2. [d]rn2k1nr/1pppp1pp/8/8/8/8/PPPPPPPP/3QK2R w Kkq - 0 1
Q vs 2Ns+R; 2 more pawns for the queen should make it about equal in terms of material. Actually, the queen side is half a pawn up still in material, so I removed the a pawn of black with some semi-open file, to possibly decrease the white advantage to 0.25cps in terms of material. Still, I think, the 3 pieces should prevail.

3. [d]rn2kb1r/1pppp1pp/8/8/8/8/PPPPPPPP/3QK2R w Kkq - 0 1
Q vs R+N+B

4. [d]r1b1kb1r/1pppp1pp/8/8/8/8/PPPPPPPP/3QK2R w Kkq - 0 1
Q vs R+2Bs; this and the previous position are good to measure the influence of the pair of bishops separately from the bonus for the 3 pieces.

5. [d]r3k1nr/1pp1p1pp/8/8/8/8/PPPPPPPP/3QK3 w kq - 0 1
Q vs 2Rs +N; the queen side has 3 pawns to level it off, the removed f pawn for black should not be a liability, as the quantity of pieces shelter the king well even in the absence of good pawn shelter. Materially again the position should be perfectly equal, but I expect the 3 pieces to somehow prevail. (but whether the 2Rs would be as efficient as 2 minors, or even more?)

6. [d]r1b1k2r/1pp1p1pp/8/8/8/8/PPPPPPPP/3QK3 w kq - 0 1
Q vs 2Rs+B;

7. [d]r1b1kb1r/1pp1p1pp/8/8/8/8/PPPPPPPP/2BQK3 w kq - 0 1
Same imbalance as in position 6, but already with black having a pair of bishops. I think again comparing this position with the previous one could be a good indication of the influence of the pair of bishops.

I think all positions are materially almost or perfectly equal. Maybe you or someone else could use them to measure more precisely the influence of the imbalance, the influence of the pair of bishops within the imbalance, as well as possible differences in the performance of various configurations of 3 pieces.

[Lyudmil Tsvetkov]

It is a bit unclear to me what you guys mean when speaking about 'handling the imbalance'. Normally imbalances 'handle themselves', because you just add piece values. So it seems there is an underlying assumption here that piece values are defined by fitting them to other imbalences than Q vs 3 minors, and that the thus obtained values are somehow no good in the case at hand. But of course Q vs 3 minors could have been used to define the Queen value.

Kaufman values (with Q=975 not the way he corrects it to 950) would have Q exactly equal to BNN (3x 325). Lyudmil is correct in remarking that against BBN the minors have an extra advantage of 50; an imbalances Q-BBN is worse for the Queen as Q-BNN.

In both cases the minors appear to have the advantage. I think that is also what common wisdom says. This would indeed suggest that 975 is an overestimate for Q, and 950 would be better.

Note that if 'elephantiasis' plays any role here, it is probably already incorporated in the base values.

Using this method, I always have difficulty in getting the correct value of Rooks and similar pieces (which are difficult to develop from behind a closed wall of Pawns). They always seem to test ~25 cP too low compared to what you would expect. Of course we all know that Rooks benefit from open files (perhaps as much as 25 cP?), and starting a Wazir (a 1-step Rook) in front of the Pawns indeed make it test 25 cP stronger than when you trap it behind the Pawns. Of course for a Rook starting it in front of the Pawns would not be very helpful, as it immediately gets harassed by minors, and might even be lost to them. It is just too valuable for that. (While a Wazir is only ~130 cP.)

Just 2 remarks:

- it would be wrong to incorporate the values obtained for pieces from this specific imbalance into the standard piece values. A very clear distinction should be made: you have an imbalance with equal number of pieces or one of the sides having just a piece more, the values obtained from this particular imbalance do not work; you have an imbalance with one side having 2 pieces more, the imbalance values work. This imbalance with 2 pieces more stands quite apart from everything else. I think the best approach would be not to change piece values, but just add an additional ad hoc bonus for the case when there are 2 pieces more.

- 2Ns+B vs Q the queen should have advantage, but in the pure case when there are no other pieces apart from those. Probably when other pieces are present, this might change the existing balance, but the question is how if at all. I do not think anyone has done any particular research on this (if yes, please share), but my feeling is that an additional piece already helps the side with the bigger number of pieces, as coordination increases. So that, while the queen is obviously better in the pure case of Q vs 2Ns + B being the only pieces, Q+R vs 2Ns+B+R might already somewhat change the balance, with the advantage of the queen side decreasing, but I do not believe the minors side could still arrive at equality.

[hgm]

Note that deleting multiple Pawns usually gives a lot of compensation, in terms of speed of development. Some of the positions you show even have a Rook on a half-open file; If Thomas alternates first move between black and white, black would immediately grab a Pawn there...

I also notice that your test positions delete extra pieces on both sides, not part of the imbalance. Not sure why you do that. Deleting an extra Knight on both sides doesn't affect the number of Bishop pairs.

Note that in general pieces do not need to have the same relative values in the opening as in the early end game. So I usually try to measure with as many pieces as possible on the board, so I can be sure that what I determine are opening values. Measuring Q-BNN with otherwise nothing but Pawns (8 each) on the board might give a different value from when there are also Rooks. (Elephantiasis predicts this, btw.)

If you put very little material on the board, the method might no longer give accurate results, because the outcome is already determined. It works best if the evaluation still has the opportunity to spread out (by imperfect play) over a range larger than the draw margin, so that it spills over both in the win and loss sector, and the total result of the match then tells you by how much. If you set up a drawish late-end-game initial position, the engines will not allow the evaluation to spread so much that you get anywhere near a win or loss, and a tiny advantage for one side or the other will remain invisible, as the result will always be draw.

[hgm]

- it would be wrong to incorporate the values obtained for pieces from this specific imbalance into the standard piece values. A very clear distinction should be made: you have an imbalance with equal number of pieces or one of the sides having just a piece more, the values obtained from this particular imbalance do not work; you have an imbalance with one side having 2 pieces more, the imbalance values work. This imbalance with 2 pieces more stands quite apart from everything else. I think the best approach would be not to change piece values, but just add an additional ad hoc bonus for the case when there are 2 pieces more.

If I understand this correctly, you say that piece values obtained by fitting empirical advantages of 1 vs 1 piece, would do a poor job in predicting 2 vs 1 imbalances, and vice versa (and similar for 3 vs 1). I never noticed a strong effect of this, when I was fitting the piece values of Capablanca Chess. The problem is that Pawn value is ill defined; Pawns are very cooperative pieces, and with hugely different numbers of Pawns one side will get many passers. So you cannot really measure the Queen in a 1 vs 1 imbalance, not even against a Rook. (Perhaps you could measure on a pure Pawn scale by introducing 3 unorthodox intermediates, such as ArchBishop (B+N) or Dragon (R+K). Then you never need more than an imbalance of 1 Pawn, and you can express all differences in terms of f2/f7 value.)

- 2Ns+B vs Q the queen should have advantage, but in the pure case when there are no other pieces apart from those. Probably when other pieces are present, this might change the existing balance, but the question is how if at all. I do not think anyone has done any particular research on this (if yes, please share), but my feeling is that an additional piece already helps the side with the bigger number of pieces, as coordination increases. So that, while the queen is obviously better in the pure case of Q vs 2Ns + B being the only pieces, Q+R vs 2Ns+B+R might already somewhat change the balance, with the advantage of the queen side decreasing, but I do not believe the minors side could still arrive at equality.

Well, elephantiasis indeed predicts that extra Rooks would be a disadvantage for the Queen, the latter having to avoid exchange for Rook, while the minors would strive for such an exchange. The effect would be partly alleviated by the fact that R-for-R trading is possible. I am not so sure that pure Q vs BNN would be an advantage for the Queen with any (equal) number of Pawns, though.

[tpetzke]

Hi,

positions with the a7 pawn removed are not so good as when I assign the first move right to black he gets an additional pawn.

Currently when I have to remove 3 pawns I remove b7, c7 and f7.

I do not intent to alter the basic material piece values, I'm planning just to scan for the imbalance and assign a bonus for the side with the better material if it is present.

I already have that for certain imbalances and now I'm adding a few more.

Thomas...

[Lyudmil Tsvetkov]

Hi,

positions with the a7 pawn removed are not so good as when I assign the first move right to black he gets an additional pawn.

Currently when I have to remove 3 pawns I remove b7, c7 and f7.

I do not intent to alter the basic material piece values, I'm planning just to scan for the imbalance and assign a bonus for the side with the better material if it is present.

I already have that for certain imbalances and now I'm adding a few more.

Thomas...

Hi Thomas. That is what I came up with.
I thought you could start all the games with white, should not be big difference.

I hope you will share the results when ready...

[Lyudmil Tsvetkov]

Note that deleting multiple Pawns usually gives a lot of compensation, in terms of speed of development. Some of the positions you show even have a Rook on a half-open file; If Thomas alternates first move between black and white, black would immediately grab a Pawn there...

I also notice that your test positions delete extra pieces on both sides, not part of the imbalance. Not sure why you do that. Deleting an extra Knight on both sides doesn't affect the number of Bishop pairs.

Note that in general pieces do not need to have the same relative values in the opening as in the early end game. So I usually try to measure with as many pieces as possible on the board, so I can be sure that what I determine are opening values. Measuring Q-BNN with otherwise nothing but Pawns (8 each) on the board might give a different value from when there are also Rooks. (Elephantiasis predicts this, btw.)

If you put very little material on the board, the method might no longer give accurate results, because the outcome is already determined. It works best if the evaluation still has the opportunity to spread out (by imperfect play) over a range larger than the draw margin, so that it spills over both in the win and loss sector, and the total result of the match then tells you by how much. If you set up a drawish late-end-game initial position, the engines will not allow the evaluation to spread so much that you get anywhere near a win or loss, and a tiny advantage for one side or the other will remain invisible, as the result will always be draw.

Make better positions and share them.

[hgm]

Hi Thomas. That is what I came up with.
I thought you could start all the games with white, should not be big difference.

The side that moves first usually has a 3-4% advantage (~ 1/6 of a Pawn), in a position where most pieces still need to be developed.