OK, I write now, when I have time, and people interested might test ideas or participate in the discussion when they have time or feel like doing so.
It would still be nice if at least one person posts into this thread, so this does not look like a single performance.
Some more pawns
Moderator: Ras
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Advanced binds
I do not know if this works or not, but is looks like a reasonable idea.
An advanced pawn bind would be any any 2 own pawns controlling simultaneously one of the 6 squares on the 7th rank: b7,c7,d7,e7,f7 or g7.
b7 square would be controlled by a6 and c6 pawns, c7 square by b6 and d6 pawns, d7 square by c6 and e6 pawns, e7 square by d6 and f6 pawns, f7 square by e6 and g6 pawns, and g7 square by f6 and h6 pawns
Such pawns could be either normal pawns, one pawn and one passer, or frequently 2 passers.
[d]6k1/p1p4p/P1P2P1P/8/8/8/8/6K1 w - - 0 1
advanced binds on the b7 and g7 squares
[d]6k1/8/1P1P4/8/8/2p1p3/8/6K1 w - - 0 1
advanced binds, formed by passers, on the c7 square for white and d2 square for black
The bonus for advanced binds on the 7th rank could be some 10-30cps, depending on what your engine has and how the terms are tuned.
The bonus should be due both for the mg and eg.
Again, this seems like a very reasonable term, but I do not know if it works.
As a rule, when one side has 2 such advanced pawns close by, the game would already be won, so the point is does this in any way help the engine see the win early enough.
As obvious, eval is meaningless, if you are able to search everything.
Eval is only useful, when it helps the engine see things in advance.
An advanced pawn bind would be any any 2 own pawns controlling simultaneously one of the 6 squares on the 7th rank: b7,c7,d7,e7,f7 or g7.
b7 square would be controlled by a6 and c6 pawns, c7 square by b6 and d6 pawns, d7 square by c6 and e6 pawns, e7 square by d6 and f6 pawns, f7 square by e6 and g6 pawns, and g7 square by f6 and h6 pawns
Such pawns could be either normal pawns, one pawn and one passer, or frequently 2 passers.
[d]6k1/p1p4p/P1P2P1P/8/8/8/8/6K1 w - - 0 1
advanced binds on the b7 and g7 squares
[d]6k1/8/1P1P4/8/8/2p1p3/8/6K1 w - - 0 1
advanced binds, formed by passers, on the c7 square for white and d2 square for black
The bonus for advanced binds on the 7th rank could be some 10-30cps, depending on what your engine has and how the terms are tuned.
The bonus should be due both for the mg and eg.
Again, this seems like a very reasonable term, but I do not know if it works.
As a rule, when one side has 2 such advanced pawns close by, the game would already be won, so the point is does this in any way help the engine see the win early enough.
As obvious, eval is meaningless, if you are able to search everything.
Eval is only useful, when it helps the engine see things in advance.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Definition for a wedge
What is a wedge?
I think people might disagree on this, but for me, there is a single reasonable definition that could be usefully applied in eval.
A wedge should be a blocked d6 or e6 pawn that is not a lever.
[d]6k1/3p4/3P4/8/8/4p3/4PP2/6K1 w - - 0 1
d6 is a white wedge above
e3 is not a black wedge, as it is a lever, meaning currently unstable feature
So, there are only 2 wedge pawns on the entire board: those are to be found on the d6 and e6 squares on the 6th rank, when the respective pawn is blocked by an enemy pawn and not a lever/attacking enemy pawn.
The requirement for the pawn to be blocked is absolutely essential, as advanced passers are already considered in eval, and besides they do not quite qualify for the wedge definition.
The condition that the pawn is not a lever is also absolutely essential.
Some people might generalise wedges as pawns on c6,d6,e6,f6 squares, and even on d5 and e5 squares, but that is very far from the truth. In the first place, c6/f6 pawns would frequently also be storming pawns, considered in eval, that would be redundant, and in the second, and most importantly, in order for a wedge to exist, you need an advanced central pawn, cutting the board in 2. Only pawns that do that are blocked e6 and d6 pawns.
The wedge bonus might be some 10-30cps, depending on what you have in your engine, due only for the mg.
Now, you can further elaborate on wedges, by specifying if the respective blcoked pawn on e6 or d6 is defended or not, twice defended, or the spearhead/most advanced pawn of a long chain.
Of course, defended pawns and especially spearheads of long chains are particularly strong, as more permanent, so you could give different bonus points in respect of that, but again, for me, in the simplest case, any blocked d6 or e6 pawn that is not a lever is a wedge and should receive the bonus.
[d]6k1/3p4/2pP4/2P3p1/5pP1/4pP2/4P3/6K1 w - - 0 1
defended white wedge on d6 - very strong
black wedge on e3 that is spearhead of a long chain - even stronger
I wonder how many people have wedges in their eval?
I think people might disagree on this, but for me, there is a single reasonable definition that could be usefully applied in eval.
A wedge should be a blocked d6 or e6 pawn that is not a lever.
[d]6k1/3p4/3P4/8/8/4p3/4PP2/6K1 w - - 0 1
d6 is a white wedge above
e3 is not a black wedge, as it is a lever, meaning currently unstable feature
So, there are only 2 wedge pawns on the entire board: those are to be found on the d6 and e6 squares on the 6th rank, when the respective pawn is blocked by an enemy pawn and not a lever/attacking enemy pawn.
The requirement for the pawn to be blocked is absolutely essential, as advanced passers are already considered in eval, and besides they do not quite qualify for the wedge definition.
The condition that the pawn is not a lever is also absolutely essential.
Some people might generalise wedges as pawns on c6,d6,e6,f6 squares, and even on d5 and e5 squares, but that is very far from the truth. In the first place, c6/f6 pawns would frequently also be storming pawns, considered in eval, that would be redundant, and in the second, and most importantly, in order for a wedge to exist, you need an advanced central pawn, cutting the board in 2. Only pawns that do that are blocked e6 and d6 pawns.
The wedge bonus might be some 10-30cps, depending on what you have in your engine, due only for the mg.
Now, you can further elaborate on wedges, by specifying if the respective blcoked pawn on e6 or d6 is defended or not, twice defended, or the spearhead/most advanced pawn of a long chain.
Of course, defended pawns and especially spearheads of long chains are particularly strong, as more permanent, so you could give different bonus points in respect of that, but again, for me, in the simplest case, any blocked d6 or e6 pawn that is not a lever is a wedge and should receive the bonus.
[d]6k1/3p4/2pP4/2P3p1/5pP1/4pP2/4P3/6K1 w - - 0 1
defended white wedge on d6 - very strong
black wedge on e3 that is spearhead of a long chain - even stronger
I wonder how many people have wedges in their eval?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
How to successfully apply backward-fated pawns
Very simple: whenever you have a backward pawn, the square in front of which is simultaneously controlled/attacked by 2 enemy pawns, you will have a backward-fated pawn.
[d]6k1/8/4p3/3p4/3P1P2/8/8/6K1 w - - 0 1
most engines consider e6 black pawn above as backward, but it is also backward-fated, as 2 enemy white pawns, d4 and f4, simultaneously control the square in front of it.
Of course, backward-fated pawns are due larger penalty than simple backward pawns, for obvious immobilisation reasons.
A very good approach to scoring backward-fated pawns would be to simply increase the backward pawn penalty by further 1/4.
In this way, wherever such a pawn is found, this will tie the cumulative penalty to the existing penalty for the particular square, which for me is the right approach.
If you do not tie the penalty to specific squares, this might certainly create a problem.
I really thought a lot about this, and for me, again, this is the right approach to implement backward-fated pawns, by simply increasing the respective penalty by 1/4 whenever the square in front of the pawn is controlled by 2 enemy pawns.
My conclusions are based on a very large sample of observed games.
[d]6k1/2p5/1p5p/1P1P2p1/2p1p1P1/2P1P3/3P4/6K1 w - - 0 1
simple h6 black backward pawn
backward-fated black c7 pawn
backward-fated white d2 pawn
Any thoughts on this?
[d]6k1/8/4p3/3p4/3P1P2/8/8/6K1 w - - 0 1
most engines consider e6 black pawn above as backward, but it is also backward-fated, as 2 enemy white pawns, d4 and f4, simultaneously control the square in front of it.
Of course, backward-fated pawns are due larger penalty than simple backward pawns, for obvious immobilisation reasons.
A very good approach to scoring backward-fated pawns would be to simply increase the backward pawn penalty by further 1/4.
In this way, wherever such a pawn is found, this will tie the cumulative penalty to the existing penalty for the particular square, which for me is the right approach.
If you do not tie the penalty to specific squares, this might certainly create a problem.
I really thought a lot about this, and for me, again, this is the right approach to implement backward-fated pawns, by simply increasing the respective penalty by 1/4 whenever the square in front of the pawn is controlled by 2 enemy pawns.
My conclusions are based on a very large sample of observed games.
[d]6k1/2p5/1p5p/1P1P2p1/2p1p1P1/2P1P3/3P4/6K1 w - - 0 1
simple h6 black backward pawn
backward-fated black c7 pawn
backward-fated white d2 pawn
Any thoughts on this?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Tripled pawns
Tripled pawns definitely have their part in eval, and you could gain, even if a small amount of elo, by correctly implementing that.
In distinction to quadrupled pawns, that are very rare, lacking practical importance, tripled pawns happen from time to time in real games.
[d]6k1/6p1/8/6p1/6p1/6P1/8/6K1 w - - 0 1
black has tripled pawns above in the face of g7,g5 and g4
Now, as far as I know, engines in general would consider any tripled pawn as simply a second doubled pawn.
For example SF does that. So, whenever you find a second doubled pawn, you give the same penalty, but the truth is that the second doubled pawn, actually the tripled pawn, is due a much higher penalty.
For me, the rule should be as this: whenever you find a doubled pawn, you give the respective penalty; whenever you find a second doubled pawn, or actually a tripled pawn, you double the doubled pawn penalty for that pawn.
So, above, if g7 is a normal pawn, g5 is doubled pawn, receiving its normal penalty, but g4 is already a tripled pawn, due twice the standard doubled pawn penalty.
The rule could be applied of course in this way: if you find a more advanced pawn on the same file, you give the normal doubled penalty; if you find a pawn that is more advanced than a doubled pawn, you give a tripled pawn penalty, equal to 2 standard penalties.
How stupid you think such a rule is?
In distinction to quadrupled pawns, that are very rare, lacking practical importance, tripled pawns happen from time to time in real games.
[d]6k1/6p1/8/6p1/6p1/6P1/8/6K1 w - - 0 1
black has tripled pawns above in the face of g7,g5 and g4
Now, as far as I know, engines in general would consider any tripled pawn as simply a second doubled pawn.
For example SF does that. So, whenever you find a second doubled pawn, you give the same penalty, but the truth is that the second doubled pawn, actually the tripled pawn, is due a much higher penalty.
For me, the rule should be as this: whenever you find a doubled pawn, you give the respective penalty; whenever you find a second doubled pawn, or actually a tripled pawn, you double the doubled pawn penalty for that pawn.
So, above, if g7 is a normal pawn, g5 is doubled pawn, receiving its normal penalty, but g4 is already a tripled pawn, due twice the standard doubled pawn penalty.
The rule could be applied of course in this way: if you find a more advanced pawn on the same file, you give the normal doubled penalty; if you find a pawn that is more advanced than a doubled pawn, you give a tripled pawn penalty, equal to 2 standard penalties.
How stupid you think such a rule is?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Doubly defended outposts
Now, this is very reasonable and significant feature, but at the same time very tricky, as you have to be extremely specific, in order to succeed with that.
[d]6k1/8/3p4/2pNp3/2P1P3/8/8/6K1 w - - 0 1
Nd5 is a doubly defended outpost above
Of course, doubly defended outposts are obviously outposts defended by 2 own pawns.
As far as I know, engine in general give additional bonus to defended outposts, but I think they make no distinction whether the outpost is defended by a single or 2 own pawns.
And there is a very major distinction here, but you have to be extremely specific.
Now, for knight outposts, it does not make sense to consider all ranks and files, but only doubly defended outposts on the 5th and 6th ranks, files c through f.
So, for knights, you should exclude outposts on the 3rd and 4th ranks, as well as outposts on the 5th and 6th ranks on the files that are not central or semicentral.
This leaves knights outposts on the c5,d5,e5,f5,c6,d6,e6,f6 squares.
For bishop outposts, only 2 squares are relevant: double defended bishop outposts on the d6 and e6 squares.
So, for bishops, 3rd,4th and 5th ranks are entirely irrelevant, and on the 6th rank just the d6 and e6 squares.
Whenever you find doubly defended minor outposts on the abovementioned squares, you might increase the available bonus by another 1/4, probably by another 1/4 over the larger bonus, that would be the bonus for the general case of an outpost defended by a pawn.
Now, there is a lot to gain from that, as outposts are an important eval feature, but you should be diligent and specific when applying it.
Giving the bonus for all outposts will simply fail.
I do not know how SF will apply this, although the engine has difficulties with the concept, as SF never submits patches that would change more than 2 lines of code, and here you must be very specific to succeed.
Only reasonable suggestion to try for SF without changing more lines of code is to consider only knight outposts on the 5th and 6th ranks, although that might be somewhat off target already, as it will include also files b and g.
For diligent programmers, who are ready to write more lines of code and be specific, I really hope the concept brings some benefit.
How many engine authors have eval for doubly defended outposts?
[d]6k1/8/3p4/2pNp3/2P1P3/8/8/6K1 w - - 0 1
Nd5 is a doubly defended outpost above
Of course, doubly defended outposts are obviously outposts defended by 2 own pawns.
As far as I know, engine in general give additional bonus to defended outposts, but I think they make no distinction whether the outpost is defended by a single or 2 own pawns.
And there is a very major distinction here, but you have to be extremely specific.
Now, for knight outposts, it does not make sense to consider all ranks and files, but only doubly defended outposts on the 5th and 6th ranks, files c through f.
So, for knights, you should exclude outposts on the 3rd and 4th ranks, as well as outposts on the 5th and 6th ranks on the files that are not central or semicentral.
This leaves knights outposts on the c5,d5,e5,f5,c6,d6,e6,f6 squares.
For bishop outposts, only 2 squares are relevant: double defended bishop outposts on the d6 and e6 squares.
So, for bishops, 3rd,4th and 5th ranks are entirely irrelevant, and on the 6th rank just the d6 and e6 squares.
Whenever you find doubly defended minor outposts on the abovementioned squares, you might increase the available bonus by another 1/4, probably by another 1/4 over the larger bonus, that would be the bonus for the general case of an outpost defended by a pawn.
Now, there is a lot to gain from that, as outposts are an important eval feature, but you should be diligent and specific when applying it.
Giving the bonus for all outposts will simply fail.
I do not know how SF will apply this, although the engine has difficulties with the concept, as SF never submits patches that would change more than 2 lines of code, and here you must be very specific to succeed.
Only reasonable suggestion to try for SF without changing more lines of code is to consider only knight outposts on the 5th and 6th ranks, although that might be somewhat off target already, as it will include also files b and g.
For diligent programmers, who are ready to write more lines of code and be specific, I really hope the concept brings some benefit.
How many engine authors have eval for doubly defended outposts?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Doubly defended outposts
I forgot to mention that the bonus for doubly defended outposts would be due just for the mg.
Of course, the main reason the bonus is due is that doubly defended outposts are even a more vital form of space advantage, not only difficult, as in the case with outposts defended by a single pawn, but altogether impossible to eliminate.
If such outposts are attacked and eliminated by enemy minors, this will leave a very strong defended pawn, quite often a passer, on either the advanced 5th, or 6th rank. Attacking and eliminating single defended outposts is much much easier, and usually does not leave strong pawns/passers on advanced ranks.
It really does not make sense to consider such outposts in the eg, the usual bonus points might suffice there.
[d]6k1/8/2pBp3/2PpPp2/3PnP2/8/8/6K1 w - - 0 1
doubly defended bishop outpost on the d6 square. Eliminating this bishop leaves behind a very strong advanced white defended passer on d6
doubly defended black knight outpost on the e4 square. Eliminating this knight leaves behind a strong passer, or at least a strong defended pawn in some cases, when there is pawn in front of the minor.
Of course, the main reason the bonus is due is that doubly defended outposts are even a more vital form of space advantage, not only difficult, as in the case with outposts defended by a single pawn, but altogether impossible to eliminate.
If such outposts are attacked and eliminated by enemy minors, this will leave a very strong defended pawn, quite often a passer, on either the advanced 5th, or 6th rank. Attacking and eliminating single defended outposts is much much easier, and usually does not leave strong pawns/passers on advanced ranks.
It really does not make sense to consider such outposts in the eg, the usual bonus points might suffice there.
[d]6k1/8/2pBp3/2PpPp2/3PnP2/8/8/6K1 w - - 0 1
doubly defended bishop outpost on the d6 square. Eliminating this bishop leaves behind a very strong advanced white defended passer on d6
doubly defended black knight outpost on the e4 square. Eliminating this knight leaves behind a strong passer, or at least a strong defended pawn in some cases, when there is pawn in front of the minor.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Doubly defended non-outpost knights
Knights that are not outposts, are also due some additional ad-hoc bonus, when defended by 2 own pawns.
The bonus might be some 5-10cps, due only for the mg.
Why is it so, is a bit difficult to exaplain now, but that is simply so.
Of course, that would be true only of knights on the 5th and 6th ranks and central and semicentral files: squares c5,d5,e5,f5,c6,d6,e6,f6.
[d]6k1/1pp2pp1/5N2/2N1P1P1/1P1P4/8/8/6K1 b - - 0 1
Although not an outpost, Nc5 is due an additional bonus above.
Same is true of Nf6 - this might help with some sacrifices, etc.
The bonus might be some 5-10cps, due only for the mg.
Why is it so, is a bit difficult to exaplain now, but that is simply so.
Of course, that would be true only of knights on the 5th and 6th ranks and central and semicentral files: squares c5,d5,e5,f5,c6,d6,e6,f6.
[d]6k1/1pp2pp1/5N2/2N1P1P1/1P1P4/8/8/6K1 b - - 0 1
Although not an outpost, Nc5 is due an additional bonus above.
Same is true of Nf6 - this might help with some sacrifices, etc.