Lever evaluation

[Lyudmil Tsvetkov]

Mr. Kmoch did a great job coming up with the lever concept.

Only thing not entirely positive about it is that he mostly suggested concepts that do not quite work, like inner/outer levers, etc.

There are things that work for levers, and things that do not work.
Just like for every other pawn.

If you have 10-20 standard pawn features, of course you can combine those in a way that they produce a whopping one, 2 or 5 thousand or more specific pawn sub-features, but the trick is to be able to combine the different features the right way, for some combinations are fully meaningless, while others should be quite productive.

Specifically for levers, one of the most important features is whether the lever is opposed or unopposed.
People have gotten accustomed to using the opposed/unopposed flag only to assess the extent of weakness of backward and isolated pawns, but of course, the flag could be used on many other occasions.

[Lyudmil Tsvetkov]

Well, of course, it makes very much sense to distinguish between levers on the 5th and 6th ranks in terms of whether they are opposed/have an enemy pawn on the same file, or unopposed/do not have an enemy pawn on the same file.

Obviously, unopposed levers are due the bigger bonus.

[d]6k1/5p2/1p4p1/2P2P2/8/8/8/6K1 w - - 0 1

c5 is an unopposed lever on the 5th rank

f5 is an opposed lever on the 5th rank

[d]6k1/pp4p1/1P3P2/8/8/8/8/6K1 w - - 0 1

b6 is an opposed lever on the 6th rank

f6 is an unopposed lever on the 6th rank

So, it makes sense to score unopposed levers 20% or so higher than opposed levers.

For example, in SF, levers on the 5th and 6th ranks are scored with 20 and 40cps bonus respectively. If an opposed flag is applied, the value for opposed levers on the 5th/6th rank could become 18/38, while the value for unopposed levers 22/42 cps. I even think the distinction is bigger, so unopposed levers might get 25/45 cps, while opposed levers 15/35 cps.

You might think this is not important, but this is not so, as those are simply different pawn features with their different chess meaning, and they should be scored accordingly.

What do you think of this idea?

[Lyudmil Tsvetkov]

Of course, although levers on the 4th rank would be symmetrical and useless to evaluate, if generally applied, adding an unopposed flag already makes unopposed levers on the 4th rank asymmetrical and worthy of scoring.

Unopposed levers on the 4th rank are due some small bonus, maybe some 5-7cps, both for the mg and eg.

[d]6k1/8/1p6/p4p2/1P2P3/P7/5P2/6K1 w - - 0 1

b4 and a5 are opposed levers on the 4th rank, that cancel each other and do not deserve any bonus.

e4 and f5 are levers on the 4th rank, but, while e4 is an unopposed lever on the 4th rank, f5 is an opposed lever on the 4th rank for black, as it is opposed by the white f2 pawn.
That makes the distinction and the e4 white lever worthy of scoring.

Of course, sometimes unopposed levers on the 4th rank cancel each other for both sides, but that is the same with all other eval features, so no problem here.

How many of you already score unopposed levers on the 4th rank?
Apart from Daniel, I mean.

[Lyudmil Tsvetkov]

Double levers - pawns attacking 2 enemy pawns at the same time - have their meaning in eval, but only on the 6th rank.

Double levers on the 5th and other ranks apart from the 6th have no real eval meaning.
It does not matter at that if such double levers on the 6th rank are opposed or unopposed, a bonus is always valid.

The bonus for a double lever on the 6th rank might be some 50% higher than the already assigned value for a general lever on the 6th rank, so if you score levers on the 6th rank with say, 40cps, you could add another 20cps over that.

[d]6k1/ppp2p1p/1P4P1/P7/8/8/8/6K1 w - - 0 1

b6 is a double lever on the 6th rank

g6 is another double lever on the 6th rank

Please do not tell me that no one scores double levers on the 6th rank.

[Lyudmil Tsvetkov]

A pair/duo of levers on the 5th and 6th ranks - 2 lever pawns on the same rank next to each other on adjacent files - are of course due an additional bonus, primarily because of tactical reasons, as is the usual case with most levers.

The overbonus might reach some 50% higher value for each lever of the pair, but you could experiment of course with lower values.

[d]6k1/5ppp/1pp3PP/1PP2P2/8/8/8/6K1 w - - 0 1

b5 and c5 are a lever pair on the 5th rank for white; g6 and h6 are a lever pair on the 6th rank for white

It is extremely difficult to neutralise tactical threats created by lever pairs, much more so than in the usual case with levers.

Please note, that if you use the concept of connected pawns as applied in SF to score lever pairs, this will lead you nowhere, as SF scores both pawns next to each other on the same rank and defended pawns as connected, in exactly the same way, but, while it makes very much sense to score lever duos, levrs next to each other on the same rank, it makes no sense at all to score differently defended levers.

So one should be very careful when applying the terms.

How much do you dislike this notion?

[Lyudmil Tsvetkov]

Just to mention again that by combining pawn features, you might be able to score not 10,20,50 different pawn features, but at least a couple of thousand.

And all those couple of thousand pawn terms will of course be a reflection of chess board reality and duly deserve some evaluation values.

The task will be to find those pawn feature combinations that really matter chesswise, and neglect the other ones.

Some might think that no one needs so many different terms, but this is not so.
A while ago I thought that SF and other top chess engines made a mistake/suboptimal move every second move, but obviously I have been wrong - my current assessment is that those top engines make suboptimal moves every single move, and one of the main reasons for that is, of course, poor evaluation.

There are so many possible chess positions, and so many different eval terms applying in them that, without any doubt, creating a 10 000 elo-strong engine is fully possible.

When do you think engines will reach the 10 000 elo mark?

[cdani]

Hi.
I had not levers on Andscacs. Now I'm testing them.

1st only basic levers at 5th and 6th ranks without any distinction. This gave 3 elo at short time control but 0 elo at long time control, so I discarded it.

Then adding the distinction if there is an adversary pawn in front or not gave may be 1-3 elo at 20000 games, so I kept it.

Then I tried adding some bonus to the pawn attacking two others, also gave 0.

Now I'm trying joined double lever. I will post.
Thanks.

[Lyudmil Tsvetkov]

Hi.
I had not levers on Andscacs. Now I'm testing them.

1st only basic levers at 5th and 6th ranks without any distinction. This gave 3 elo at short time control but 0 elo at long time control, so I discarded it.

Then adding the distinction if there is an adversary pawn in front or not gave may be 1-3 elo at 20000 games, so I kept it.

Then I tried adding some bonus to the pawn attacking two others, also gave 0.

Now I'm trying joined double lever. I will post.
Thanks.

Thanks for the feedback, Jose!

Please do not forget to try unopposed lever on the 4th rank, might be promising.

With Andscacs, everything is easy, the problem is how to improve SF...

[cdani]

Please do not forget to try unopposed lever on the 4th rank, might be promising.

It was already inside the patch of unopposed levers, so was a little good.

[Lyudmil Tsvetkov]

Please do not forget to try unopposed lever on the 4th rank, might be promising.

It was already inside the patch of unopposed levers, so was a little good.

OK, but the size of the bonus for the 4th rank should be very small, maybe 1/4 of that for the 5th rank.

Also, when trying it make sense to try separately, not to mingle with higher ranks. AT least what I see on the board is that such levers have some meaning chesswise...

Thanks again for the feedback.