Binders

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Binders

Post by Lyudmil Tsvetkov » Wed Feb 04, 2015 9:21 pm

If I were Hans Kmoch, I could coin my very strange-sounding name for a specific chess eval feature.

Imagining myself for a moment in Hans's skin, binders would be any 2 pawns, controlling one and the same square on the board.

For example, b3 and d3 white pawns would be binders, as they simultaneously control the c4 square; for the same reason, h4 and f4 would be binders, as they simultaneously control the g5 square.

Binders are very important eval feature, as double pawn control is a prominent characteristic of a wide range of positions, which sometimes can drastically change the eval.

However, as mentioned, the key to good, better and excellent evaluation is to apply any concept precisely to the squares where it belongs.
In this particular case, I think binders are meaningful only in a very limited number of squares on the board, spanning the 5th and 6th ranks, as well as files c through f.

So if you ask me how to evaluate binders, I would limit them to the c5, d5, e5, f5 squares on the 5th rank, as well as the c6, d6, e6 and f6 squares on the 6th rank.

So any 2 pawns controlling at the same time the above 8 squares should receive some bonus. I would give some 30cps in the general case, but maybe the 5th rank can get 25cps, while the 6th rank 35cps.

So, b4 and d4 pawns would receive bonus for controlling c5 square at the same time; d4 adn f4 pawns for controlling e5 square at the same time; d5 and f5 pawns for controlling the e6 square at the same time; e5 adn g5 pawns for controlling the f6 square at the same time, etc.

So binders are the 2 pawns controlling the abovementioned squares, but we count a single bonus just for squares controlled, the 8 specific squares on the 5th and 6th ranks in the above scenario.

Someone might think that that is another stupid concept, however here I would vouch that this is very important and largely underestimated by engines of all strength.

For example, thanks to such pawns, openings like the Dutch Stonewall, the Stonewall Attack for white (the reverse of the Dutch Stonewall), the Maroczy Bind in the Sicilian, etc., are fuly sound, promising and vital, though many engines and some players might not quite agree.

Below some examples:

[d]r2q1rk1/1ppb1ppp/p1np1n2/2b1p3/2P1P3/2NP2PP/PP2NPBK/R1BQ1R2 b - - 0 1

This is some special kind of bind, I frequently play in my games against SF.
I call it the English Bind, as it usually arises after the moves c4 e5 Nc3 Nc6 g3 Nf6 Bg2 Bb4 e4 or c4 e5 Nc3 Nf6 e4

It is also possible to get this position via a transposition of the Sicilian, e4 c5 c4, and that is one of the reasons I consider the Sicilian not fully equal for black, because white can play c4 on the second move, but who would agree?

Never mind, the above position strongly favours white, no matter what engines might think.
White threatens f4-f5, with a strong attack, and the peculiarity of the position is that white firmly controls the center at the same time with the 2 white binders on c4 and e4, so black does not have counterplay in the center, can not play d6-d5 for a very long time, almost for ever, while the counterplay on the distant queen side is not as satisfactory and rapid.

It is also possible that white altogether wins this, but that would mean that both c4 and e4 win for white. :)

So, as binders c4 and e4 pawns above deserve some nice bonus, at least some 30cps, but how many engines would see it like that?

[d]rnbq1rk1/1p2b1pp/p1p1pn2/3p1p2/2PP1B2/2NBPN2/PP3PPP/R2Q1RK1 b - - 0 1

Another example.
A typical position of the Dutch Stonewall.

No matter what different engines might think, here black has at least full equality, and I even think black is better.

Although black lags seriously behind in development, 2 main eval feature turn the tables in black's favour:

- the d5 apex pawn, which is very strong and cements the position
- and the d5 and f5 binders, controlling the important e4 square; on this vital square later can land a black knight and stay there for a very long time, as it is supported not by one, but 2 own pawns

So I would give some nice bonus for the apex d5 pawn above, as well as the f5 and d5 binders, controlling the e4 square.

[d]r4rk1/1bb1qp1p/4p1p1/1p1pPnP1/p1pP2Q1/P1P1PR2/1PBN3P/R1B3K1 w - - 0 1

One last example.

e5 and g5 binders controlling the important f6 square definitely deserve a nice bonus; the same would not be true of the c4 and a4 black binders, controlling the b3 square, as this square is not that important and can be safely skipped.

Lastly, a distinction should be made between an apex pawn, a pawn defended by 2 own pawns, and a binder, a square controlled by 2 own pawns.

In the first case, you have another own pawn on the destination square, and in the second case, with binders, it really does not matter what you have on the doubly controlled destination square, it is usually empty, but there might also be a friendly piece or an enemy pawn. In the case of binders, it also does not quite matter if the opponent also has a pawn or 2 controlling the same square, though it is of course preferable for the opponent to control it as insufficiently as possible.

So with binders, you might just specify a square controlled by 2 own pawns.
Also, in distinction to apex pawns, where the important squares are on the 4th and 5th ranks, files c through f, with binders the important squares are on the 5th and 6th ranks, same files.

What do you think of this another stupid idea of mine?
Anyone implementing something similar?

Dicaste
Posts: 85
Joined: Mon Apr 16, 2012 5:23 pm
Location: Istanbul, TURKEY

Re: Binders

Post by Dicaste » Wed Feb 04, 2015 9:46 pm

I am not an expert about evaluation positions but i agree with your claims Lyudmil but other way. I play exactly same style like yours unlikely classical chess. I play chess like a real war. Do you really use your important soldiers first ? or try to keep them for another/important situation ?. I like to keep. I like to exchange knights for have a bishop for an endgame example. Like i said i play chess like real war. Pawns first!! Pawns should be on forward and knights/bishops should be behind on them for me. Important thing is once you move pawn you can't take back to early position. I really don't know binders actually i like your ideas always but the key of your position is pawns on forward minor pieces behind them. Maybe thats why binders are really important.

carldaman
Posts: 1630
Joined: Sat Jun 02, 2012 12:13 am

Re: Binders

Post by carldaman » Wed Feb 04, 2015 11:18 pm

Hi Lyudmil,

I really like the concept. Perhaps the reason engines don't 'like' the Stonewall formations is because they don't understand binders -- they may even undervalue or penalize them.

The only drawback to consider is the (usually) empty square between the binders, which can become a weakness in one's camp where an enemy piece could land or control from a distance. Even with this drawback, binders should be a net positive factor in eval.

Regards,
CL
Lyudmil Tsvetkov wrote:If I were Hans Kmoch, I could coin my very strange-sounding name for a specific chess eval feature.

Imagining myself for a moment in Hans's skin, binders would be any 2 pawns, controlling one and the same square on the board.

For example, b3 and d3 white pawns would be binders, as they simultaneously control the c4 square; for the same reason, h4 and f4 would be binders, as they simultaneously control the g5 square.

Binders are very important eval feature, as double pawn control is a prominent characteristic of a wide range of positions, which sometimes can drastically change the eval.

However, as mentioned, the key to good, better and excellent evaluation is to apply any concept precisely to the squares where it belongs.
In this particular case, I think binders are meaningful only in a very limited number of squares on the board, spanning the 5th and 6th ranks, as well as files c through f.

So if you ask me how to evaluate binders, I would limit them to the c5, d5, e5, f5 squares on the 5th rank, as well as the c6, d6, e6 and f6 squares on the 6th rank.

So any 2 pawns controlling at the same time the above 8 squares should receive some bonus. I would give some 30cps in the general case, but maybe the 5th rank can get 25cps, while the 6th rank 35cps.

So, b4 and d4 pawns would receive bonus for controlling c5 square at the same time; d4 adn f4 pawns for controlling e5 square at the same time; d5 and f5 pawns for controlling the e6 square at the same time; e5 adn g5 pawns for controlling the f6 square at the same time, etc.

So binders are the 2 pawns controlling the abovementioned squares, but we count a single bonus just for squares controlled, the 8 specific squares on the 5th and 6th ranks in the above scenario.

Someone might think that that is another stupid concept, however here I would vouch that this is very important and largely underestimated by engines of all strength.

For example, thanks to such pawns, openings like the Dutch Stonewall, the Stonewall Attack for white (the reverse of the Dutch Stonewall), the Maroczy Bind in the Sicilian, etc., are fuly sound, promising and vital, though many engines and some players might not quite agree.

Below some examples:

[d]r2q1rk1/1ppb1ppp/p1np1n2/2b1p3/2P1P3/2NP2PP/PP2NPBK/R1BQ1R2 b - - 0 1

This is some special kind of bind, I frequently play in my games against SF.
I call it the English Bind, as it usually arises after the moves c4 e5 Nc3 Nc6 g3 Nf6 Bg2 Bb4 e4 or c4 e5 Nc3 Nf6 e4

It is also possible to get this position via a transposition of the Sicilian, e4 c5 c4, and that is one of the reasons I consider the Sicilian not fully equal for black, because white can play c4 on the second move, but who would agree?

Never mind, the above position strongly favours white, no matter what engines might think.
White threatens f4-f5, with a strong attack, and the peculiarity of the position is that white firmly controls the center at the same time with the 2 white binders on c4 and e4, so black does not have counterplay in the center, can not play d6-d5 for a very long time, almost for ever, while the counterplay on the distant queen side is not as satisfactory and rapid.

It is also possible that white altogether wins this, but that would mean that both c4 and e4 win for white. :)

So, as binders c4 and e4 pawns above deserve some nice bonus, at least some 30cps, but how many engines would see it like that?

[d]rnbq1rk1/1p2b1pp/p1p1pn2/3p1p2/2PP1B2/2NBPN2/PP3PPP/R2Q1RK1 b - - 0 1

Another example.
A typical position of the Dutch Stonewall.

No matter what different engines might think, here black has at least full equality, and I even think black is better.

Although black lags seriously behind in development, 2 main eval feature turn the tables in black's favour:

- the d5 apex pawn, which is very strong and cements the position
- and the d5 and f5 binders, controlling the important e4 square; on this vital square later can land a black knight and stay there for a very long time, as it is supported not by one, but 2 own pawns

So I would give some nice bonus for the apex d5 pawn above, as well as the f5 and d5 binders, controlling the e4 square.

[d]r4rk1/1bb1qp1p/4p1p1/1p1pPnP1/p1pP2Q1/P1P1PR2/1PBN3P/R1B3K1 w - - 0 1

One last example.

e5 and g5 binders controlling the important f6 square definitely deserve a nice bonus; the same would not be true of the c4 and a4 black binders, controlling the b3 square, as this square is not that important and can be safely skipped.

Lastly, a distinction should be made between an apex pawn, a pawn defended by 2 own pawns, and a binder, a square controlled by 2 own pawns.

In the first case, you have another own pawn on the destination square, and in the second case, with binders, it really does not matter what you have on the doubly controlled destination square, it is usually empty, but there might also be a friendly piece or an enemy pawn. In the case of binders, it also does not quite matter if the opponent also has a pawn or 2 controlling the same square, though it is of course preferable for the opponent to control it as insufficiently as possible.

So with binders, you might just specify a square controlled by 2 own pawns.
Also, in distinction to apex pawns, where the important squares are on the 4th and 5th ranks, files c through f, with binders the important squares are on the 5th and 6th ranks, same files.

What do you think of this another stupid idea of mine?
Anyone implementing something similar?

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: Binders

Post by Lyudmil Tsvetkov » Thu Feb 05, 2015 12:02 pm

Dicaste wrote:I am not an expert about evaluation positions but i agree with your claims Lyudmil but other way. I play exactly same style like yours unlikely classical chess. I play chess like a real war. Do you really use your important soldiers first ? or try to keep them for another/important situation ?. I like to keep. I like to exchange knights for have a bishop for an endgame example. Like i said i play chess like real war. Pawns first!! Pawns should be on forward and knights/bishops should be behind on them for me. Important thing is once you move pawn you can't take back to early position. I really don't know binders actually i like your ideas always but the key of your position is pawns on forward minor pieces behind them. Maybe thats why binders are really important.
Great to have a supporter from Istambul. :)

There are thousand possible and reasonable eval parameters, and it is just a matter of tuning.

I do not know what is preferable: pawns first, or pieces first - it would depend on the specific position and the specific set of available eval terms. Sometimes pawns first helps, sometimes hurts.

I mainly post positions engines do not understand.

Those pawns attack/control important squares of the board in a sound way, it is good to have control over those squares, so for me this feature is meaningful. It is another matter however to tune such an element within the framework of an engine's eval and search...

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: Binders

Post by Lyudmil Tsvetkov » Thu Feb 05, 2015 12:16 pm

carldaman wrote:Hi Lyudmil,

I really like the concept. Perhaps the reason engines don't 'like' the Stonewall formations is because they don't understand binders -- they may even undervalue or penalize them.

The only drawback to consider is the (usually) empty square between the binders, which can become a weakness in one's camp where an enemy piece could land or control from a distance. Even with this drawback, binders should be a net positive factor in eval.

Regards,
CL
Hi Carl.

It seems you are more in a hurry than me. :)

Yes, you are fully to the point: the hole between the 2 binders is the negative element in the binders approach, and it is a matter of playing style to successfully use the binders and minimise the consequences of the hole.

Though, here I should mention 2 things why the hole is relatively less important:

- the binders would squeeze any enemy piece or pawn landed on this hole, and when 2 pawns squeeze an enemy object, this is usually bad, though slightly, for the squeezed object

- I mentioned just 8 important squares on the board, half of which are deep on the 6th rank in the opponent's camp, so 2 binders on the 5th rank, g5 and e5 for example, would leave a hole on f5, deep into the opponent's camp, that could certainly more or less be neglected

And so on, and so forth...

User avatar
asanjuan
Posts: 210
Joined: Thu Sep 01, 2011 3:38 pm
Location: Seville, Spain

Re: Binders

Post by asanjuan » Thu Feb 05, 2015 3:00 pm

Lyudmil Tsvetkov wrote:
carldaman wrote:Hi Lyudmil,

I really like the concept. Perhaps the reason engines don't 'like' the Stonewall formations is because they don't understand binders -- they may even undervalue or penalize them.

The only drawback to consider is the (usually) empty square between the binders, which can become a weakness in one's camp where an enemy piece could land or control from a distance. Even with this drawback, binders should be a net positive factor in eval.

Regards,
CL
Hi Carl.

It seems you are more in a hurry than me. :)

Yes, you are fully to the point: the hole between the 2 binders is the negative element in the binders approach, and it is a matter of playing style to successfully use the binders and minimise the consequences of the hole.

Though, here I should mention 2 things why the hole is relatively less important:

- the binders would squeeze any enemy piece or pawn landed on this hole, and when 2 pawns squeeze an enemy object, this is usually bad, though slightly, for the squeezed object

- I mentioned just 8 important squares on the board, half of which are deep on the 6th rank in the opponent's camp, so 2 binders on the 5th rank, g5 and e5 for example, would leave a hole on f5, deep into the opponent's camp, that could certainly more or less be neglected

And so on, and so forth...
As far as I know, the topic of the position is the understanding of strong and weak squares.
Having 2 binders is useless if you can't reach the strong square with a piece (specially a knight). The only thing that you objectively have is a weakness in the intermediate square.
I think that it is better to evaluate something like the occupancy of those strong squares, the control of the strong squares with our pieces and the defence of our own weak square. Maybe is a more general concept than the binders.

This is what I'm doing in Rhetoric.

By the way, the bonus of controlling the weak squares is about 4 cp per each square or so... nothing very high, and it's been automatically tunned.

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: Binders

Post by Lyudmil Tsvetkov » Thu Feb 05, 2015 9:59 pm

asanjuan wrote:
Lyudmil Tsvetkov wrote:
carldaman wrote:Hi Lyudmil,

I really like the concept. Perhaps the reason engines don't 'like' the Stonewall formations is because they don't understand binders -- they may even undervalue or penalize them.

The only drawback to consider is the (usually) empty square between the binders, which can become a weakness in one's camp where an enemy piece could land or control from a distance. Even with this drawback, binders should be a net positive factor in eval.

Regards,
CL
Hi Carl.

It seems you are more in a hurry than me. :)

Yes, you are fully to the point: the hole between the 2 binders is the negative element in the binders approach, and it is a matter of playing style to successfully use the binders and minimise the consequences of the hole.

Though, here I should mention 2 things why the hole is relatively less important:

- the binders would squeeze any enemy piece or pawn landed on this hole, and when 2 pawns squeeze an enemy object, this is usually bad, though slightly, for the squeezed object

- I mentioned just 8 important squares on the board, half of which are deep on the 6th rank in the opponent's camp, so 2 binders on the 5th rank, g5 and e5 for example, would leave a hole on f5, deep into the opponent's camp, that could certainly more or less be neglected

And so on, and so forth...
As far as I know, the topic of the position is the understanding of strong and weak squares.
Having 2 binders is useless if you can't reach the strong square with a piece (specially a knight). The only thing that you objectively have is a weakness in the intermediate square.
I think that it is better to evaluate something like the occupancy of those strong squares, the control of the strong squares with our pieces and the defence of our own weak square. Maybe is a more general concept than the binders.

This is what I'm doing in Rhetoric.

By the way, the bonus of controlling the weak squares is about 4 cp per each square or so... nothing very high, and it's been automatically tunned.
Weak squares are one thing, double/binding control over important squares another. Each of them has its meaning.

Please note that while weak squares occur on the 3rd and 4th ranks, binders control will occur on the 5th and 6th ranks, so those are quite different things.

Weak squares of course deserve some penalty, own pieces controlling those weak squares, presumably minors, some small bonus, for example yours 4 cps, while binding control deserves some nice bonus.

You are not right, binders are not useless, unless you have a minor piece to land there, quite the opposite. The point of binders is to control some important squares on the board, independently of whether an own minor will land there, if the squares will be defended by an enemy pawn at the same time, or defendable, quite in contrast to weak squares, where you would require that those weak squares/holes are undefended and undefendable by own pawns.

Binders presume control over the squares themselves, solely out of positional purposes.

Please take a look at the below 2 examples:

[d]r2q1rk1/1ppb1ppp/p1np1n2/2b1p3/2P1P3/2NP2PP/PP2NPBK/R1BQ1R2 b - - 0 1

This one I posted yesterday.

white has advantage here

d4 is a weak square/hole in the white position, on the 4th rank, and of course should be penalised; also opponent pieces can land there; but, tis penalty is frequent and much less significant than the bonus the binders c4 and e4 control over the d5 square deserves.

Please note, the d5 square is on the 5th rank, it is defendable by the c7 pawn if it goes to c6, but still the bonus is due. Why? Because d5 is an important break for black and currently it is very difficult to do it; you will need a lot of time to play c6, d5, etc., while the control will still remain.

So while white can easily play f4 attacking on the king side, because of the binders control over the d5 square black can not find sufficient counterplay in the center with d5, and this is a standard break. So c4 and e4 binders are actually very useful. You do not care if an own piece can land there or not. You prevent counterplay, and that is an extremely important positional element.

Another example

[d]r1bqnrk1/1p2npbp/p2p2p1/2pPp3/2P1P1PP/2NBBP2/PP1QN3/2KR3R b - - 0 1

Typical position of the Saemish Variation in the King's Indian, where white plays f3 and castles long away from a dangerous black chain.

White has advantage here. The Saemish is possibly the strongest and most efficient reply to black's KID and possibly the only line to seriously question the KID setup.

Why white plays f3? Because it indends to play g4 afterwards to stop the black f7-f7 break.

Look very carefully at this position.
g4 and e4 white pawns are supposed to be binders, firmly controlling the important f5 square.
Please note, that no piece lands on f5, and the f5 square is simultaneously defended by an enemy black g6 pawn, but that does not matter much and does not deprive the white binders of their usefulness in the least.

The g4 and e4 pawns prevent black to play the standard f7-f5 break with ease, and possibly black will never be able to do so. Upon f7-f5, the g file is opened and that is dangerous. Besides, even if white captures twice on f5, it can place a piece on the important e4 square, something that would otherwise be impossible to do.
If white has only one pawn on e4 and no pawn on g4, but on g2 instead, black easily plays f7-f5, and as the exhange ef gf is bad for white, losing the center, the f pawn goes further to f4, where it is very strong. The binders prevent precisely that. So they are very useful just controlling specific squares and doing nothing else.

The squares I mentioned on the 5th rank: c5,d5,e5 and f5, are all central and semicentral squares, where the opponent will most likely look for counterplay. When you have binders controlling those squares, the black counterplay in the center with e7-e5/e6-e5, d7-d5/d6-d5, or on the flanks with f7-f5 and c7-c5/c6-c5 will be thwarted, and those are the most important breaks, much more so than say b7-b5 or g5.

Similarly for the 6th rank.

So binders are actually an extremely important eval term, dealing with strictly positional play. Probably precisely because of that engines will have hard time understanding and implementing the concept; or, shall we say it the other way: precisely because of their ignorance of binders and similar positional concepts engines have hard time understanding positional play and making positional moves.

Anyone looking for a better grasp of how binds are played might want to refer to some Smyslov games.

User avatar
Evert
Posts: 2920
Joined: Fri Jan 21, 2011 11:42 pm
Location: NL
Contact:

Re: Binders

Post by Evert » Fri Feb 06, 2015 1:11 pm

Lyudmil Tsvetkov wrote: Binders are very important eval feature, as double pawn control is a prominent characteristic of a wide range of positions, which sometimes can drastically change the eval.

However, as mentioned, the key to good, better and excellent evaluation is to apply any concept precisely to the squares where it belongs.
In this particular case, I think binders are meaningful only in a very limited number of squares on the board, spanning the 5th and 6th ranks, as well as files c through f.

So if you ask me how to evaluate binders, I would limit them to the c5, d5, e5, f5 squares on the 5th rank, as well as the c6, d6, e6 and f6 squares on the 6th rank.

So any 2 pawns controlling at the same time the above 8 squares should receive some bonus. I would give some 30cps in the general case, but maybe the 5th rank can get 25cps, while the 6th rank 35cps.
Here's a potential problem associated with scoring a feature like this: say you implement this. The engine sees an opportunity to create these "binders" and does so. Now what?

One of the purposes of the evaluation is to give the engine a sense of progress towards a goal (winning the game). Advancing passers does this, attacking the king position does this and increasing mobility does this (because odds are that if you have many options, at least one of them will lead to a position where you can make progress). So how does this term help the engine make progress? Say, I have the "binders". What do I do next? Put a piece on the controlled square (I guess it would count as an outpost and get a bonus for that)? Increase pressure on the enemy pawn structure?

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: Binders

Post by Lyudmil Tsvetkov » Fri Feb 06, 2015 3:51 pm

Evert wrote:
Lyudmil Tsvetkov wrote: Binders are very important eval feature, as double pawn control is a prominent characteristic of a wide range of positions, which sometimes can drastically change the eval.

However, as mentioned, the key to good, better and excellent evaluation is to apply any concept precisely to the squares where it belongs.
In this particular case, I think binders are meaningful only in a very limited number of squares on the board, spanning the 5th and 6th ranks, as well as files c through f.

So if you ask me how to evaluate binders, I would limit them to the c5, d5, e5, f5 squares on the 5th rank, as well as the c6, d6, e6 and f6 squares on the 6th rank.

So any 2 pawns controlling at the same time the above 8 squares should receive some bonus. I would give some 30cps in the general case, but maybe the 5th rank can get 25cps, while the 6th rank 35cps.
Here's a potential problem associated with scoring a feature like this: say you implement this. The engine sees an opportunity to create these "binders" and does so. Now what?

One of the purposes of the evaluation is to give the engine a sense of progress towards a goal (winning the game). Advancing passers does this, attacking the king position does this and increasing mobility does this (because odds are that if you have many options, at least one of them will lead to a position where you can make progress). So how does this term help the engine make progress? Say, I have the "binders". What do I do next? Put a piece on the controlled square (I guess it would count as an outpost and get a bonus for that)? Increase pressure on the enemy pawn structure?
Very difficult question.

It is a matter of tuning of eval and search terms.

So far none of the strictly positional eval features I have suggested has passed successful tests and been implemented, as far as I know.
Neither backward-fated pawns, nor apex pawns, nor bimders, nor KID structures, nor using the knowledge of existing closed sides on the board.

But those are the real positional elements that all top human grandmasters possess, and even only stronger players.
They are long-term elements and are a vital feature of high-quality chess.

Now, for me, it is not a matter of depth, as engines search very deep in general; it is not also a matter of wrong eval, as the features are real and tuning is always possible.

For me, it is a problem of engine search being specifically tuned to play more open and less positional chess. I do not know the exact terms, but say, you tune futility pruning, LMR, different search extensions, etc., to work precisely with the set of existing eval terms, and as the existing eval terms in most engines are standard regular features that have nothing to do with positionality, those other subtle positional elements I mentioned above simply fail the test.

So, in order for them to succeed, you need a new approach, based on tuning newly introduced positional eval terms with the whole set of existing seacrh parameters.

That is how I see it, but I might be wrong, never mind.

I can not add much more about binders: below is a game I just played against latest SF 6.

[pgn][PlyCount "95"]
[Event "Blitz 5m+3s"]
[Site "Sofia"]
[Date "2015.02.06"]
[White "Tsvetkov, Lyudmil"]
[Black "Stockfish 6 64 POPCNT"]
[Result "1-0"]
[ECO "D00"]
[TimeControl "300+3"]
[Annotator "Tsvetkov,Lyudmil"]
[MLNrOfMoves "47"]
[MLFlags "000100"]

{512MB, Dell XPS 4Cores} 1. f4 {[%emt 0:00:00]} 1... Nf6 {[%emt 0:00:16]} 2.
Nf3 {[%emt 0:00:04]} 2... e6 {[%emt 0:00:01]} 3. d4 {[%emt 0:00:06]} 3... d5
{[%emt 0:00:09]} 4. e3 {[%emt 0:00:04]} 4... c5 {[%emt 0:00:06]} 5. c3
{[%emt 0:00:02]} 5... Be7 {[%emt 0: 00:18]} 6. Bd3 {[%emt 0:00:06]} 6... O-O
{[%emt 0:00:31]} 7. O-O {[%emt 0:00:02]} 7... c4 {[%emt 0:00:10]} 8. Bc2
{[%emt 0:00:02]} 8... b5 {[%emt 0:00:12]} 9. a3 {[%emt 0:00: 04]} 9... Bb7
{[%emt 0:00:08]} 10. Ne5 {[%emt 0:00:06]} 10... a5 {[%emt 0:00:10]} 11. Nd2
{[%emt 0:00:04]} 11... a4 {[%emt 0:00:02]} 12. Qf3 {[%emt 0:00:05]} 12... Nc6
{[%emt 0:00: 09]} 13. g4 {[%emt 0:00:05]} 13... g6 {[%emt 0:00:19]} 14. g5
{[%emt 0:00:02]} 14... Nxe5 {[%emt 0:00:07]} 15. dxe5 {[%emt 0:00:10]} 15... Nd7
{[%emt 0:00:00]} 16. Qg4 {[%emt 0:00:10]} 16... Nc5 {[%emt 0:00:13]} 17. Nf3
{[%emt 0:00:04]} 17... Qb6 {[%emt 0:00:39]} 18. Nd4 {[%emt 0:00:14]} 18... Ne4
{[%emt 0:00:13]} 19. Rf3 {[%emt 0:00:24]} 19... Bc6 {[%emt 0:00:14]} 20. Rh3
{[%emt 0:00:06]} 20... Rfc8 {[%emt 0:00:24]} 21. Qe2 {[%emt 0:00:14]} 21... Qd8
{[%emt 0:00:08]} 22. Bd2 {[%emt 0:00:06]} 22... Qf8 {[%emt 0:00:21]} 23. Kh1
{[%emt 0:00:24]} 23... Nc5 {[%emt 0:00:06]} 24. Rg1 {[%emt 0:00:28]} 24... Ne4
{[%emt 0:00:34]} 25. Rg4 {[%emt 0:01:19]} 25... Qg7 {[%emt 0:00:00]} 26. Be1
{[%emt 0: 00:42]} 26... Qh8 {[%emt 0:00:00]} 27. Rgh4 {[%emt 0:00:26]} 27... Bd7
{[%emt 0:00:00]} 28. Bd1 {[%emt 0:00:19]} 28... Rd8 {[%emt 0:00:09]} 29. Qg2
{[%emt 0:00:11]} 29... Rac8 {[%emt 0:00:06]} 30. Rh6 {[%emt 0:00:09]} 30... Bf8
{[%emt 0:00:07]} 31. R6h4 {[%emt 0:00: 02]} 31... Be7 {[%emt 0:00:05]} 32. Kg1
{[%emt 0:00:10]} 32... Nc5 {[%emt 0:00:23]} 33. Bc2 {[%emt 0:00:03]} 33... Rb8
{[%emt 0:00:10]} 34. Bd2 {[%emt 0:00:03]} 34... h5 {[%emt 0:00: 05]} 35. Bd1
{[%emt 0:00:02]} 35... Qg7 {[%emt 0:00:04]} 36. Bxh5 {[%emt 0:00:04]} 36... b4
{[%emt 0:00:03]} 37. axb4 {[%emt 0:00:03]} 37... Nd3 {[%emt 0:00:08]} 38. Be2
{[%emt 0:00:05]} 38... a3 {[%emt 0:00:05]} 39. bxa3 {[%emt 0:00:02]} 39... Bc5
{[%emt 0:00:01]} 40. Rh6 {[%emt 0:00:05]} 40... Bxd4 {[%emt 0:00:03]} 41. exd4
{[%emt 0:00:01]} 41... Ra8 {[%emt 0:00:02]} 42. Qg4 {[%emt 0:00:04]} 42... Rxa3
{[%emt 0:00:03]} 43. Qh4 {[%emt 0:00:02]} 43... Kf8 {[%emt 0:00:03]} 44. Rh7
{[%emt 0:00:04]} 44... Qg8 {[%emt 0:00:00]} 45. Rh8 {[%emt 0:00:02]} 45... Ke7
{[%emt 0:00:04]} 46. Rxg8 {[%emt 0:00:02]} 46... Rxg8 {[%emt 0:00:02]} 47. Qh7
{[%emt 0:00:02]} 47... Rf8 {[%emt 0:00:06]} 48. Bxd3 {[%emt 0:00:02]} 1-0
[/pgn]

Hardware was 4x1.7
5 min+ 3 sec. increment blitz
(and this time I applied strictly the time control, so I did not even think longer than SF)
ponder on
512 Mb hash

Look at some positions:

[d]rn1q1rk1/pb2bppp/4pn2/1p1p4/2pP1P2/P1P1PN2/1PB3PP/RNBQ1RK1 w - - 0 10
Important e5 bind, with f4 and d4 binders creating it

[d]r4rk1/1b2bp1p/1q2p1p1/1pnpP1P1/p1p2PQ1/P1P1PN2/1PB4P/R1B2RK1 w - - 0 18
Important f6 bind, with g5 and e5 binders creating it.
Black also has 2 binds, on f5 and b3, but they are unimportant, or less so, as the f5 bind is in the own half of the board, on the 4th rank, while the b3 bind is on the flank.

So white has significant advantage in terms of binds.

[d]1r1r2k1/3b1pq1/4p1p1/2bpP1P1/1PpN1P1R/P1PnP2R/3BB1QP/6K1 w - - 0 40

Can you believe SF, supposedly the strongest engine in the world, plays black?

Nothing more to add, just look carefully at the game and how the respective binds mainly brought white the victory.

User avatar
Evert
Posts: 2920
Joined: Fri Jan 21, 2011 11:42 pm
Location: NL
Contact:

Re: Binders

Post by Evert » Fri Feb 06, 2015 4:24 pm

Lyudmil Tsvetkov wrote: Very difficult question.

It is a matter of tuning of eval and search terms.
I suspect it's probably more complicated than that. It is sufficient to just add the evaluation term if aiming for a position with that term makes you statistically more likely to find a way to make progress. If it doesn't, it just leads to a dead-end and doesn't help to add Elo.

Let me clarify where I'm coming from. We'll need to make an excursion into chess-variant territory, specifically Shogi.

In Shogi, an important strategy is to build a castle for the king. There are a number of ways to do this, and several different strategies to attack different castle formations. My chess variant program plays Shogi, but it knows nothing about building good castles, it just tries to defend the area around its king as best it can.

Here's where it gets interesting: HGM once told me that strong Shogi programs will build "proper" castles, and will follow the correct strategy to attack (and dismantle) properly build castles. However, if you don't build a proper castle, they will penalise you for doing a shoddy job (by giving themselves a healthy positional bonus), and subsequently fail to exploit that weakness effectively. In other words, they have an evaluation term that tells them that their position is better, but they lack an evaluation term that allows them to make progress.

Now, Chess is not Shogi, in a way chess is far more material driven: if you can eventually win a piece, or simplify the game, the trading gradient will indicate that you have made progress to winning the game. Nevertheless, a similar situation can occur.

Two examples: if you award a bonus for having a passed pawn, but not for advancing it, the program will produce passed pawns and tries to hang on to them, but it will not advance them unless it can find a promotion within the search horizon.
Second example: if an engine doesn't know how to make progress in KBNK, it will not find the mate. Even KRK is difficult to win if you don't know to drive the lone king to a corner.
So far none of the strictly positional eval features I have suggested has passed successful tests and been implemented, as far as I know.
Neither backward-fated pawns, nor apex pawns, nor bimders, nor KID structures, nor using the knowledge of existing closed sides on the board.

But those are the real positional elements that all top human grandmasters possess, and even only stronger players.
They are long-term elements and are a vital feature of high-quality chess.
Right, which could have to do with the point I made above: it's not enough to have a term that tells the engine that a particular positional feature is good or bad, it also needs to have the knowledge to help it make progress.

So perhaps your ideas would be more successful if you could add something from that angle? In other words: I suspect the problem you cite with play in closed positions is that you need to formulate an evaluation term that allows the engine to make progress in a closed position, rather than just terms that allows it to evaluate it more accurately.
I can not add much more about binders: below is a game I just played against latest SF 6.
Perhaps you could annotate the game and indicate what you do to make progress?
Nothing more to add, just look carefully at the game and how the respective binds mainly brought white the victory.
You're more qualified to do that than I am, I'm afraid.

Post Reply