Double levers

[Lyudmil Tsvetkov]

Well, here is an idea how you can do levers across ranks.
You can do double levers (a pawn attacking 2 enemy pawns at the same time) across ranks, as the symmetry is broken here by the fact that one friendly pawn attacks 2 enemy pawns. So it is not a symmetrical 1vs 1 attack that could be scored just on ranks 5 and 6.

Double levers could be done on ranks 3,4,5 and 6. Rank 2 is irrelevant.
The bonus points for levers could be the following:

3rd rank - 5cps
4th rank - 10cps
5th rank - 15cps
6th rank - 30cps

For SF, if someone is interested in testing, this would translate to 10,20,30,60cps for ranks 3-6.

Below some diagrams:

[d]6k1/8/8/8/p1p5/1P6/P1P5/6K1 w - - 0 1
double b3 lever on the 3rd rank

[d]6k1/8/4n3/p1p5/1P6/P7/3B4/6K1 w - - 0 1
double b4 lever on the 4th rank

[d]5rk1/8/p1p5/1P6/2P5/8/8/1R4K1 w - - 0 1
double b5 lever on the 5th rank

[d]6k1/5ppp/6P1/5P2/8/8/8/6K1 w - - 0 1
double g6 lever on the 6th rank

In all the cases, double levers are extremely useful and efficient. They are actually much more efficient than single levers. Besides, this will help your engine gain in agressiveness.

What do you think of double levers?

PS. Sorry about posting again, I am just about to quit, but ideas start terrorising me again.

[rap]

Double levers are quite rare. Some bench examples:
[d]rnbqkb1r/ppp1pppp/5P2/3p4/8/8/PPPP1PPP/RNBQKBNR b KQkq - 0 3
The pawn will be captured
[d]rnbqkb1r/ppp1pppp/3P1n2/8/8/8/PP1PPPPP/RNBQKBNR b KQkq - 0 3
[d]rnbqkbnr/p1p1pppp/8/1p1p4/Q1P5/5N2/PP1PPPPP/RNB1KB1R w KQkq - 0 3
[d]rnbqkbnr/ppp1pppp/8/3p4/2P1P3/8/PP1P1PPP/RNBQKBNR b KQkq - 0 2
[d]rnbqk2r/pp3ppp/4pn2/1Npp4/1P1P1B2/8/PP2PPPP/R2QKBNR b KQkq - 0 6

Double lever is also attacked twice, so it might be even weaker than regular lever.

[Lyudmil Tsvetkov]

Double levers are quite rare. Some bench examples:
[d]rnbqkb1r/ppp1pppp/5P2/3p4/8/8/PPPP1PPP/RNBQKBNR b KQkq - 0 3
The pawn will be captured
[d]rnbqkb1r/ppp1pppp/3P1n2/8/8/8/PP1PPPPP/RNBQKBNR b KQkq - 0 3
[d]rnbqkbnr/p1p1pppp/8/1p1p4/Q1P5/5N2/PP1PPPPP/RNB1KB1R w KQkq - 0 3
[d]rnbqkbnr/ppp1pppp/8/3p4/2P1P3/8/PP1P1PPP/RNBQKBNR b KQkq - 0 2
[d]rnbqk2r/pp3ppp/4pn2/1Npp4/1P1P1B2/8/PP2PPPP/R2QKBNR b KQkq - 0 6

Double lever is also attacked twice, so it might be even weaker than regular lever.

Double levers are rarer, but much more efficient. I have no doubts about that.

Usually, when a lever is attacked twice, it will also be supported by an own pawn, which makes it strong enough. On the other hand, it is much more difficult to neutralise a double attack.

I do not know how this will perform, but I think it is worth trying.

Reuven, I hope you will not curse me, but I would like to ask you for a personal favour. I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%. Just the middlegame value, not the endgame. Would it be possible for you to submit such a patch? I would have asked Joerg, but he is not around these days, and Arjun will certainly decline.

So please, would it be possible for you to do me this enormous favour to see how it performs? I would have done it myself, but I am unable to push patches.

[zullil]

I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%.
... to see how it performs?

Haven't these values already been optimized by extensive testing? If so, then what you propose will certainly fail, if tested in the same manner.

A more interesting experiment might be to test the tweaked and untweaked versions of Stockfish against another engine, Komodo for example.

[rap]

Reuven, I hope you will not curse me, but I would like to ask you for a personal favour. I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%. Just the middlegame value, not the endgame. Would it be possible for you to submit such a patch? I would have asked Joerg, but he is not around these days, and Arjun will certainly decline.

So please, would it be possible for you to do me this enormous favour to see how it performs? I would have done it myself, but I am unable to push patches.

Well, I sent a test. KnightValueMg increased from 817 to 833, which is close to BishopValueMg(836). Bishop considered better than knight.

[Lyudmil Tsvetkov]

Reuven, I hope you will not curse me, but I would like to ask you for a personal favour. I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%. Just the middlegame value, not the endgame. Would it be possible for you to submit such a patch? I would have asked Joerg, but he is not around these days, and Arjun will certainly decline.

So please, would it be possible for you to do me this enormous favour to see how it performs? I would have done it myself, but I am unable to push patches.

Well, I sent a test. KnightValueMg increased from 817 to 833, which is close to BishopValueMg(836). Bishop considered better than knight.

Many thanks, Reuven!

You are incredible!!

Now I almost feel like I have been able to push a patch myself.

[Lyudmil Tsvetkov]

I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%.
... to see how it performs?

Haven't these values already been optimized by extensive testing? If so, then what you propose will certainly fail, if tested in the same manner.

A more interesting experiment might be to test the tweaked and untweaked versions of Stockfish against another engine, Komodo for example.

No, Louis.

They have not tested changing piece values for a veeery long time.

Lastly, about half an year ago, they tried to change imbalance piece values, but that is a different thing. One disadvantage of imbalance is that you can not specify there different middlegame and endgame values.

[zullil]

I would very much like to see tested in the framework a patch increasing the middlegame value of the knight by 2%.
... to see how it performs?

Haven't these values already been optimized by extensive testing? If so, then what you propose will certainly fail, if tested in the same manner.

A more interesting experiment might be to test the tweaked and untweaked versions of Stockfish against another engine, Komodo for example.

They have not tested changing piece values for a veeery long time.

I wonder why not? I assumed these values were optimized at one point, and are checked upon regularly as the engine continues to develop.

In any case, your experiment is already in the queue. I'm guessing it will fail, but let's see!

[Uri Blass]

I think that maybe reducing the middle game value of the knight value by 2% may give a better result if the result of big elo loss of 4 rating points hold
with more games.

stockfish see advantage for black in the following position from my game that is not correct

r1bq1rk1/p3nppp/1pn1p3/3pP3/1P1N4/P1B5/2P2PPP/R2QKB1R w KQ - 1 11

[Lyudmil Tsvetkov]

I think that maybe reducing the middle game value of the knight value by 2% may give a better result if the result of big elo loss of 4 rating points hold
with more games.

stockfish see advantage for black in the following position from my game that is not correct

r1bq1rk1/p3nppp/1pn1p3/3pP3/1P1N4/P1B5/2P2PPP/R2QKB1R w KQ - 1 11

Hi Uri.

I saw so many games where SF underestimates knights in the mg preferring instead bishop, pair of bishops, pawns, queen, etc. A really astounding number of games where that would be true.

And yet the increase seems to fail currently convincingly. Some things are really difficult for me to explain in SF. I am sure a lot could be gained by tweaking piece values in SF, but, if a sensible approach does not work, I do not know what will.