Page 1 of 1

Doubled and Backward Pawn Engine "Definitions"

Posted: Mon Sep 07, 2009 2:54 pm
by brianr
Been working on pawn eval for Tinker for some time and was wondering about a couple of definitions.

Backwards:
Do you count as backwards a pawn on its starting position home rank?
Does the stop square have to be not controlled by the opponent?
Or both stop squares, when applicable?
But not if passed?

Doubled:
There are many definitions, and in some cases they may even be "good" (as in not isolated and lead pawn supported and some center influence).

My questions involve the ranks between doubled pawns.
One can perhaps minimally only count doubled pawns on adjacent ranks,
or those with just empty ranks,
or with just its own pieces in between,
or the opponent's,
or various combinations.

Initital testing with Tinker (which has some cummulative and potentially hefty doubled pawn penalties) indicates that it plays better when only counting doubled pawns on adjacent ranks.

So, just wondering what other have found.
Thanks,
Brian

Some links:

http://home.comcast.net/~danheisman/Art ... _pawns.htm

http://en.wikipedia.org/wiki/Chess_piece_relative_value

http://chessprogramming.wikispaces.com/Doubled+Pawn

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Mon Sep 07, 2009 3:10 pm
by bob
brianr wrote:Been working on pawn eval for Tinker for some time and was wondering about a couple of definitions.

Backwards:
Do you count as backwards a pawn on its starting position home rank?
Does the stop square have to be not controlled by the opponent?
Or both stop squares, when applicable?
But not if passed?
I test for the concept of a weak pawn, rather than a "backward" pawn. A pawn is weak if (a) it can not be defended by a neighbouring pawn. That doesn't mean that a neighbouring pawn is not present, but that this pawn simply can not be pushed to defend the pawn being evaluated. It might be too far advanced, or it might be unable to advance because of opponent pawns. (b) if it can't be defended by a pawn on adjacent files, then the question is, can the pawn advance to the point where it is defended by an adjacent pawn? Presently these two questions do not consider enemy pieces, only the pawns for both sides.


Doubled:
There are many definitions, and in some cases they may even be "good" (as in not isolated and lead pawn supported and some center influence).

My questions involve the ranks between doubled pawns.
One can perhaps minimally only count doubled pawns on adjacent ranks,
or those with just empty ranks,
or with just its own pieces in between,
or the opponent's,
or various combinations.

Initital testing with Tinker (which has some cummulative and potentially hefty doubled pawn penalties) indicates that it plays better when only counting doubled pawns on adjacent ranks.
I don't follow the "doubled on adjacent ranks". Do you mean (say) g2 and g3 both have pawns and they are doubled, while g2 and g4 would not be? If there are two on a file, I count 'em as doubled.


Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Mon Sep 07, 2009 3:52 pm
by brianr
bob wrote:I don't follow the "doubled on adjacent ranks". Do you mean (say) g2 and g3 both have pawns and they are doubled, while g2 and g4 would not be? If there are two on a file, I count 'em as doubled.
That is what I have always done, until recently.

I'm thinking of a somewhat reduced penalty with squares in between.

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Tue Sep 08, 2009 3:14 am
by tano-urayoan
bob wrote:I test for the concept of a weak pawn, rather than a "backward" pawn. A pawn is weak if (a) it can not be defended by a neighbouring pawn. That doesn't mean that a neighbouring pawn is not present, but that this pawn simply can not be pushed to defend the pawn being evaluated. It might be too far advanced, or it might be unable to advance because of opponent pawns. (b) if it can't be defended by a pawn on adjacent files, then the question is, can the pawn advance to the point where it is defended by an adjacent pawn? Presently these two questions do not consider enemy pieces, only the pawns for both sides.
How you differentiate between a weak pawn and a passed pawn? Example a passed pawn could be in both of your definitions for weak pawns, is this pawn strong or weak for your evaluation? How you differentiate between a weak pawn, a weak passed pawn and a strong passed pawn( in this example is just a solitary passed pawn, yet strong in the position)

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Tue Sep 08, 2009 8:49 pm
by bob
tano-urayoan wrote:
bob wrote:I test for the concept of a weak pawn, rather than a "backward" pawn. A pawn is weak if (a) it can not be defended by a neighbouring pawn. That doesn't mean that a neighbouring pawn is not present, but that this pawn simply can not be pushed to defend the pawn being evaluated. It might be too far advanced, or it might be unable to advance because of opponent pawns. (b) if it can't be defended by a pawn on adjacent files, then the question is, can the pawn advance to the point where it is defended by an adjacent pawn? Presently these two questions do not consider enemy pieces, only the pawns for both sides.
How you differentiate between a weak pawn and a passed pawn? Example a passed pawn could be in both of your definitions for weak pawns, is this pawn strong or weak for your evaluation? How you differentiate between a weak pawn, a weak passed pawn and a strong passed pawn( in this example is just a solitary passed pawn, yet strong in the position)
An isolated passed pawn is, by definition, weak. But since it is passed, the more it advances, the more the advanced passed pawn score adds in. But given the choice of a weak on non-weak passer, I'd take the non-weak passer if all else is equal. The "strong passed pawn idea" is a different animal. A passed pawn is better when a rook supports it from the rear, it is worse if it is blockaded by an enemy piece. Etc. All of those ideas are factored in. And any/all of them can easily make the pawn score quite large, even though the base pawn itself is weak.

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Wed Sep 09, 2009 6:44 pm
by Gerd Isenberg
brianr wrote:Been working on pawn eval for Tinker for some time and was wondering about a couple of definitions.

Backwards:
Do you count as backwards a pawn on its starting position home rank?
Does the stop square have to be not controlled by the opponent?
Or both stop squares, when applicable?
But not if passed?

Doubled:
There are many definitions, and in some cases they may even be "good" (as in not isolated and lead pawn supported and some center influence).

My questions involve the ranks between doubled pawns.
One can perhaps minimally only count doubled pawns on adjacent ranks,
or those with just empty ranks,
or with just its own pieces in between,
or the opponent's,
or various combinations.

Initital testing with Tinker (which has some cummulative and potentially hefty doubled pawn penalties) indicates that it plays better when only counting doubled pawns on adjacent ranks.

So, just wondering what other have found.
Thanks,
Brian
One may try different penalties for pawns, member of own stop squares, member of the own front- or back-span, or even in the case of triple pawns for pawns members of the own front- and back-span. The intersection of own front- and back-span are also the in-between squares of doubled pawns. I think those squares are fine to become occupied by own light pieces, if the front double pawn guards opponent pawn attacks. However, those inter-spans are usually not the squares own rook likes to occupy. For opponent pieces things are a bit different, specially if the square is not member of own pawn attack-spans, but still a rook has a very reduced mobility on that file.

The definitions of backwardness(1): the stop square is/can no longer defended by own pawns - that is not member of own attack-front spans, but the stop square is attacked by an opponent pawn. Some give extra penalty if stop is attacked by two opponent pawns (2) - even if that introduces a hole between the two opponent pawns (which might be considered elsewhere). If the pawns are open (3), that is not member of the opponent front-span, they may receive another penalty, specially a classical straggler on the second or third rank, scaled by number of opponent heavy pieces.

After a first pass in determining backward pawns as above, as well as rams, one may use a second pass, but considering only attack-spans of "movable" pawns, to determine additional backward pawns (4), which were only defendable in the first pass by blocked or backward pawns. Of course, one has to distinguish a straggler from a sneaker, where a lever-sac enforces a passer. Pushing backwards (center) pawns is often an strategical goal in various openings, e.g. d6 with the e4:e5 ram with the d6->d5xe4 lever option, so it probably makes sense to specially consider blocks and piece attacks of the stop square of those "strategical" stragglers. On the other hand, encouraging rooks to defend weak pawns in the endgame leads to passive play, if a lever can not be enforced.

Since isolanis and backwardness are usually defined as disjoint property from chess players point of view, the above setwise definition of backwardness holds also for isolated pawns, which might be considered by distinct subsets with different piece square tables, thus an isolani is weaker if open and stop is attacked by opponent pawn. Doubled pawns are less penalized nowadays, maybe only a few centipawns, except they share isolated and backward properties as well.

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Fri Sep 11, 2009 2:22 pm
by brianr
Gerd Isenberg wrote: One may try different penalties for pawns, member of own stop squares, member of the own front- or back-span, or even in the case of triple pawns for pawns members of the own front- and back-span. The intersection of own front- and back-span are also the in-between squares of doubled pawns. I think those squares are fine to become occupied by own light pieces, if the front double pawn guards opponent pawn attacks. However, those inter-spans are usually not the squares own rook likes to occupy. For opponent pieces things are a bit different, specially if the square is not member of own pawn attack-spans, but still a rook has a very reduced mobility on that file.

The definitions of backwardness(1): the stop square is/can no longer defended by own pawns - that is not member of own attack-front spans, but the stop square is attacked by an opponent pawn. Some give extra penalty if stop is attacked by two opponent pawns (2) - even if that introduces a hole between the two opponent pawns (which might be considered elsewhere). If the pawns are open (3), that is not member of the opponent front-span, they may receive another penalty, specially a classical straggler on the second or third rank, scaled by number of opponent heavy pieces.

After a first pass in determining backward pawns as above, as well as rams, one may use a second pass, but considering only attack-spans of "movable" pawns, to determine additional backward pawns (4), which were only defendable in the first pass by blocked or backward pawns. Of course, one has to distinguish a straggler from a sneaker, where a lever-sac enforces a passer. Pushing backwards (center) pawns is often an strategical goal in various openings, e.g. d6 with the e4:e5 ram with the d6->d5xe4 lever option, so it probably makes sense to specially consider blocks and piece attacks of the stop square of those "strategical" stragglers. On the other hand, encouraging rooks to defend weak pawns in the endgame leads to passive play, if a lever can not be enforced.

Since isolanis and backwardness are usually defined as disjoint property from chess players point of view, the above setwise definition of backwardness holds also for isolated pawns, which might be considered by distinct subsets with different piece square tables, thus an isolani is weaker if open and stop is attacked by opponent pawn. Doubled pawns are less penalized nowadays, maybe only a few centipawns, except they share isolated and backward properties as well.
Thanks for this thoughtful reply.
I understand about 1/3rd of it.
Perhaps you could post something along these lines here
http://chessprogramming.wikispaces.com/
with some diagrams, if you are so inclined.
Respectfully,
Brian

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Fri Sep 11, 2009 3:41 pm
by Gerd Isenberg
brianr wrote: Thanks for this thoughtful reply.
I understand about 1/3rd of it.
Perhaps you could post something along these lines here
http://chessprogramming.wikispaces.com/
with some diagrams, if you are so inclined.
Respectfully,
Brian
Hi Brian,
You will find some set-wise pawn stuff already in Pawn Pattern and Properties, see for instance Fills and Spans and Based on Spans, f.i.

Pawn Spans

Code: Select all

white frontspans    black rearspans
1 1 1 . . 1 1 1     1 1 1 1 . 1 1 1
1 1 1 . . 1 1 1     1 . 1 1 . . . 1
1 1 1 . . 1 1 1     . . . 1 . . . .
1 1 1 . . 1 1 1     . . . . . . . .
. 1 1 . . . 1 1  ^  . . . . . . . .
. 1 1 . . . 1 1  |  . . . . . . . .
. . . . . . . .     . . . . . . . .
. . . . . . . .     . . . . . . . .
               north
white pawns         black pawns
. . . . . . . .     . . . . . . . .
. . . . . . . .     . 1 . . . 1 1 .
. . . . . . . .     1 . 1 . . . . 1
. . . . . . . .     . . . 1 . . . .
1 . . . . 1 . .     . . . . . . . .
. . 1 . . . . .     . . . . . . . .
. 1 1 . . . 1 1     . . . . . . . .
. . . . . . . .     . . . . . . . .
               south
white rearspans     black frontspans
. . . . . . . .  |  . . . . . . . .
. . . . . . . .  v  . . . . . . . .
. . . . . . . .     . 1 . . . 1 1 .
. . . . . . . .     1 1 1 . . 1 1 1
. . . . . . . .     1 1 1 1 . 1 1 1
1 . . . . 1 . .     1 1 1 1 . 1 1 1
1 . 1 . . 1 . .     1 1 1 1 . 1 1 1
1 1 1 . . 1 1 1     1 1 1 1 . 1 1 1
Attack Spans

Code: Select all

white attack        white attack     
frontspan           rearspan         
. . 1 . 1 . . .     . . . . . . . . 
. . 1 . 1 . . .     . . . . . . . . 
. . 1 . 1 . . .     . . . . . . . . 
. . 1 . 1 . . .     . . . . . . . . 
. . . w . . . .     . . 1 w 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
 
black attack        black attack     
frontspan           rearspan         
. . . . . . . .     . . 1 . 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
. . . . . . . .     . . 1 . 1 . . . 
. . . b . . . .     . . 1 b 1 . . . 
. . 1 . 1 . . .     . . . . . . . . 
. . 1 . 1 . . .     . . . . . . . . 
. . 1 . 1 . . .     . . . . . . . . 
Feel free to ask further questions, or suggestions to explain stuff better.

Gerd

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Sun Sep 13, 2009 8:24 pm
by BubbaTough
Gerd Isenberg wrote: Hi Brian,
You will find some set-wise pawn stuff already in Pawn Pattern and Properties, see for instance Fills and Spans and Based on Spans, f.i.

Gerd
Hi Gerd, I liked a lot of the material at the site you posted. A question though. When talking about backward pawns, it seems to refer to the "stop square", which does not seems quite enough. Say for example you have a structure like white pawns at h3,g5, and black pawn at g6. The pawn at h3 to should be backward but is not as defined on this site right? Same is true but worse if white pawns at a5,b3,c5 and black pawns at a6 and b6...the pawn at b3 should be backward but its not.

-Sam

Re: Doubled and Backward Pawn Engine "Definitions"

Posted: Sun Sep 13, 2009 10:13 pm
by Gerd Isenberg
BubbaTough wrote:
Gerd Isenberg wrote: Hi Brian,
You will find some set-wise pawn stuff already in Pawn Pattern and Properties, see for instance Fills and Spans and Based on Spans, f.i.

Gerd
Hi Gerd, I liked a lot of the material at the site you posted. A question though. When talking about backward pawns, it seems to refer to the "stop square", which does not seems quite enough. Say for example you have a structure like white pawns at h3,g5, and black pawn at g6. The pawn at h3 to should be backward but is not as defined on this site right? Same is true but worse if white pawns at a5,b3,c5 and black pawns at a6 and b6...the pawn at b3 should be backward but its not.

-Sam
Hi Sam,
yes, good point. One may either apply one further test for the second next stop for so far movable pawns, which should be sufficient.
Alternatively, one may consider all opponent pawn attacks, not member of own front-attackspans as stop or tele-stop squares, to fill them down for a "backward pawn" area.

Gerd