Assorted tweaks

[Lyudmil Tsvetkov]

Actually, when I now look at this thread, I already CAN NOT follow it.

Anyone can??

Roger, have you seen those textbooks many American universities order? They weigh tons, I mean at least 2-3kgs. They are larger than a 19-inch monitor and weigh kgs. You can barely carry them. I think this is the upper reasonable limit for a textbook.

If you publish a textbook 10 kgs heavy and the size of a kitchen table, no one will read it, as impossible.

Same with threads.
You can not have a thread having more than 10 subthreads. Well, maybe, if you deliberately want to look silly and ugly, while being the author, you might expand it to 20 subthreads, but never more, simply no one will follow it anymore.

So, when I post couple of more subthreads, I will have to open another thread, or?

I very much hope you revise your decision and, as long as there are no number of posted threads per person restrictions, and, as long I have not violated the Charter of this forum in any way, simply allow me to post according to my discretion further - rest assured I will never cross the reasonable limit.

Otherwise, I think you are simply violating the Charter of this forum.

[Lyudmil Tsvetkov]

Well, that is the subthread I wanted to post earlier.

It is about long chains again.
Maybe Pawel can post here, if he has interest.

As said, the missing link to successfully evaluating long chains (more than 1 defended pawn) is precisely scoring the inner defended pawns of a long chain also in terms of whether the respective pawns of the chain are defended by pawns on less central files than the defended pawn itself, or pawns on more central files than the defended pawn.

This will decide if the long chain points towards or away from the center, and that is one of the most important qualities of long chains. If you do not consider this rule sufficiently, you will hardly succeed with long chains.

As noted in the past, the second most important thing to consider for long chains is an array for all inner pawns, based on files and ranks, with central files getting more value and more advanced ranks getting more value.

If we have to split the importance between the 2 factors, it is either 50-50, but I think the factor which pawn supports an inner pawn of a long chain, more central or less central than the inner pawn itself, is even weightier in long chain evaluation, so it is maybe 60-40.

[d]6k1/2p5/1p6/p7/2P5/1P6/P7/6K1 w - - 0 1
2 long chains with same inner pawn, b3/b6, having precisely the same characteristics. You consider it in terms of ranks and files, and both inner pawns are on the 3rd rank on the b file.

Does that mean that the bonus for both long chains is the same?

Of course not, the white chain is due a bigger bonus, as it points towards the center, while the black chain with the same inner pawn points away from the center.
In this case, the white chain is due bigger bonus, i.e. the inner b3 pawn is due bigger bonus than the b6 inner pawn.

If we score just files and ranks for inner pawns, we never get that.

If we score also the factor of whether an inner pawn is defended by a more or less central pawn, we already are doing quite nice with long chains pointing towards the center.

When we give 5cps more bonus to b3 pawn being defended by a pawn on less central file, a2, and another 5cps bonus for c4 pawn being defended by a less central pawn, b3, the overall score for the white long chain as pointing towards the center is already 10cps higher.

In distinction, black b6 does not get a bonus, as it is defended by a pawn on more central file, c7; a5 also does not get a bonus, as it is also defended by a pawn on more central file than the file where it is, b6.

[d]6k1/3p4/2p5/1p6/3P4/2P5/1P6/6K1 w - - 0 1
similarly, evaluating file of defending pawns scores the white long chain higher than the black long chain, although they have same inner pawn, c3/c6.

d4 and c3 defended pawns of the white long chain are both defended by pawns on less central files, c3 and b2 respectively, while b5 and c6 pawns are both defended by pawns on more central files, so no bonus here.

So, the white chain scores some 10cps higher, and indeed, it is quite stronger.

[d]6k1/3p4/2p5/1p2P3/p4P2/6P1/7P/6K1 w - - 0 1
2 long chains of 4 pawns each

Some engines, like SF for example, even do not consider files, so SF will think (I wonder if someone from the SF team will now appear to expressly prohibit me from posting any contents about SF, SF losing, etc., even mentioning the name, maybe Louis ) both chains deserve precisely the same bonus!

The white chain has one pawn on the 5th rank, one pawn on the 4th rank and one pawn on the 3rd rank that are defended. Fully the same the black chain.

So Sf will score those chains as equal.

When, in actual fact, the white chain is much stronger, as it points towards the center, while the black chain away from the center.
The fact that the white chain has more advanced central pawns does not change much, as in SF for example central pawns get just a minimal bonus in psqt. Besides, you should consider the chain as whole, and not a random assembly of separate pawns, so this makes the need for appropriate assessment even more urgent.

Even if you have file distinction for inner pawns, that does not change much, as both chains have almost the same inner pawns in terms of files, g and f files against b and c files.

When you score however also the defended pawns in terms of whether the pawns defending them are on less or more central files, the white chain scores much higher, as all of its 3 defended pawns are defended by less central pawn - g3 by h2, f4 by g3 and e5 by f4. So, you have 3 times 5cps, 15cps additional bonus for the white chain.

For the black chain, you have no bonus at all, as all of its defended pawns are defended by pawns on more central files - c6 by d7, b5 by c6 and a4 by b5.

Indeed, the white chain, pointing towards the center is much much stronger.

So again, if you want to successfully evaluate long chains, there are 2 obligatory elements:

- bonus for the inner pawns of the long chain in terms of files and ranks, very small bonus, but necessary to be distinguished between files and ranks, and

- bonus for all defended pawns of the chain in terms of whether those defended pawns are defended by own pawns on less or more central files

I think without those 2 you will go nowhere, both are almost equally important.

Pointed chains, chains pointed towards the enemy king, are quite a separate thing, and should be scored apart. Actually, in many cases, the general bonus for long chains according to the above 2 elements and the specific bonus for a chain pointed towards the enemy king will come into contradiction, but chains pointed towards the enemy king represent just a very small part of the overall percentage of existing long chains, so this could be somehow tuned, or specified, making both features working as, although rare, chains pointed at the enemy king are a weighty eval factor.

If someone is interested in long chains, he might post here, possibly have a discussion, unless something extraordinary happens.

[Lyudmil Tsvetkov]

If I see correctly that the 2 threads have been separated again, meaning I will have the discretion to post some more messages, but not cross the reasonable limits, many thanks, Roger!

A liberal approach is always the best way to proceed.

[Werewolf]

[quote="Lyudmil Tsvetkov"]

[d]6k1/5p2/5Pp1/6Pp/7P/8/8/6K1 w - - 0 1
the black king is safe, no open fies to attack it

I would disagree with this. This is one of the weaker black positions.

[Joerg Oster]

[d]6k1/5p2/5Pp1/6Pp/7P/8/8/6K1 w - - 0 1
the black king is safe, no open fies to attack it

I would disagree with this. This is one of the weaker black positions.

OTOH, the white king is completely without pawn shelter.
It depends on the location of the other pieces, of course, but in general I would prefer Black here.

[Lyudmil Tsvetkov]

Well, this is certainly not the way to moderate a forum.

Not a single word of reply either from Roger, or from any other of the moderators.

Yesterday I thought Roger would have unlocked the other thread, but he did not.

What does that mean, anyone can explain?

I will try to post some more subthreads here, but, with the best of my will, I will be unable to post more than 5 new subthreads without losing the thread of the topic completely.

So, when I post, if I do, another 5 subthreads, I will open another eval thread.

Roger, you are warned, whether you reply or not.

The thing I most hate in the world is bureaucracy.
Now I can not think about ideas, but I should think about some threads.

You and the other mods, who did not intervene, make me really feel desolate and unhappy, very desolate and very unhappy. I even do not know if I will have the force to post even a single new thing, so unhappy I feel.

Obviously, the goal is to make me leave this forum for good and completely, which I will probably do.

Only thing that still keeps me here are my nice memories of my friends here, Louis, Joerg, Carl, Ralph, and the many thanks I got during the years from different programmers.

Now, I know what has happened, someone has complained to the point that the mods simply can not proceed otherwise, but do something to punish me.

I still expect a word of response from some of the moderators, especially Roger, who took the decision.

I am really very, very disappointed.

[Lyudmil Tsvetkov]

Well, again, those are the 2 elements that long chains need:

1. scoring in terms of rank and file for inner pawns of long chains, below is my proposed array for b3-g3-g6-b6 inner pawns

Code: Select all

5   8  12  12  8  5                    g3 f3 e3 d3 c3 b3 
 10 15 22 22 15 10                   g4 f4 e4 d4 c4 b4 
 15 22 33 33 22 15                   g5 f5 e5 d5 c5 b5 
 20 30 45 45 30 20                   g6 f6 e6 d6 c6 b6 
 

or, you could also try with half of the above values, I posted the 2 arrays also in a recent thread about long chains, the subthread is called Why long chains do not work, you might check the 2 arrays there.

and 2. bonus for defended pawns of the long chain for the pawns that are defended by less central pawns than the defended pawn itself.

So, you give an additional 5cps bonus, both for the mg and the eg, whenever a pawn on a file defends a pawn on b file, a pawn on b file defends a pawn on c file, fro short I will enumerate just the files where the pawns are, so you give the bonus, when:

- a defends b, or b defends c, or c defends d, or h defends g, or g defends f, or f defends e; in the other cases, you do not give the bonus, also not when e defends d or d defends e

I think this is not so difficult to implement, actually very straightforward. But, for SF, that will not work (man, how I am afraid Joerg or Louis will appear now, or even Marco in persona, to tell me - Do not ever mention again SF does something wrong or is able to lose!) , it will not work, as SF like to write only 2 lines of code, not more, and here you have to specify that:

1. inner pawns of long chains are scored both in terms of files and ranks, and
2. all defended pawns, especially defended pawns of long chains, get additional small bonus, when they are defended by a pawn on less central file, meaning, the chain is pointed towards the center and not the rim of the board

The abovementioned bonus should be applicable to all defended pawns, but it is especially important for long chains, as there the whole chain as a single unit is affected by that rule.

[Lyudmil Tsvetkov]

Well, maybe this has been mentioned in the past, but certainly it was not devoted a specific subthread.

Outposts are very important, as you have 4 minor pieces, more than half of all pieces, and you should take good care of them.

There might exist at least some 20-50 different types of minor outposts, depending on whether they are defended by pawns, number of defending pawns, specific ranks, whether the ouposts are attackable by enemy pawns or not, to what extent they are not attackable, etc.

Scoring just outposts that are not attackable by enemy pawns is not enough.

There is a nother kind of semioutposts, that are potentially attackable by enemy pawns, but very difficult to do so, because of specific reasons, so you might almost consider them as a special kind of outposts. Leaving just to psqt to care for all the specific cases is obviously very unsatisfactory decision.

Very briefly, what would those semioutposts involve:

1. minor piece that is attackable by enemy pawns, but the pawn that could attack it is blocked by an own pawn

[d]6k1/7p/3p2pP/2pP2B1/2N5/8/8/6K1 w - - 0 1
Nc4 is semioutposted, the only enemy pawn that could attack it, d6, is blocked by white d5 pawn

same for Bg5, the only enemy pawn that could attack it, h7, is blocked by white h6 pawn

In this case, you could score 1/2 the value of the respective minor outpost

2. minor piece that is attackable by enemy pawns, but the pawn that could attack it is backward

[d]6k1/1p6/p4p2/P1N3p1/4B1P1/8/8/6K1 w - - 0 1
Nc5 is semioutposted, the only enemy pawn that could attack it, b7, is backward

same for Be4, the only enemy pawn that could attack it, f6, is backward

In this case, you could score 1/3 the value of the respective minor outpost over psqt

3. minor piece that is attackable by enemy pawns, but the square in front of the enemy pawn is controlled/attacked by an own pawn

[d]6k1/1pp5/4p3/1P3p2/1N6/4B1P1/8/6K1 w - - 0 1

Nb4 is semioutposted, the square in front of the only enemy pawn that could attack it, c7, is controlled by an own pawn, b5, i.e. b5 controls/attacks the c6 square

above, some engines score minor behind pawn, but that only partially sees specific cases of semioutposted pieces, while caring also for other unrelated issues

same for Be3, semioutposted, as the square in front of the only enemy pawn that could attack it, f5, is controlled by an own pawn, g3

In this case, you could score 1/4 to 1/8 of the respective outposts over psqt

All the best in scoring semioutposts, as the benefit here is huge, some semioutposts are almost full outposts, but of course, you will need so much tuning and untuning of existing outpost and psqt values, that I am completely at a loss how those changes could succeed, although being very reasonable.

[Lyudmil Tsvetkov]

Well, if there is one main doubled pawn rule apart from the general penalty for any doubled pawn, that will be precisely the case with overpenalty for symmetrical doubled opposed pawns.

Now, I do not know if it makes sense to score doubled pawns in terms of opposed and unopposed, however, the opposed flag is very useful for the specific case of symmetrical doubled opposed pawns.

What is a symmterical doubled opposed pawn?

Any doubled pawn that is opposed, and the 2 adjacent files are both symmetrical in terms of pawn placements, you have no pawns at all on those files, or each side has one pawn for each file.

Symmetry means it is much more difficult to undouble such pawns, semiopen files left open by doubled pawns are more difficult to exploit, etc.

[d]6k1/1p3pp1/2p4p/8/8/1P4P1/1PP2PP1/6K1 w - - 0 1
b3 is symmetrical doubled opposed pawn:
- on the adjacent a file there are no pawn for both sides
- on the adjacent c files there is one own and one enemy pawn

g3 is not a symmetrical doubled pawn:
- while on the adjacent f file there is one own and one enemy pawn
- on the adjacent h file there is just one enemy pawn, so this breaks the symmetry

b3 is a bigger weakness of course, as more difficult to undouble; g3 is easier to undouble, or you can make use of the semiopen h file

[d]6k1/pp3pp1/2p1p2p/2P5/5P2/P7/1PP2PPP/6K1 w - - 0 1
symmetrical c5 doubled opposed pawn
- d files is open, no own or enemy pawns
- on b file you have one own and one enemy pawn

asymmetrical f4 doubled pawn:
- on g file you have one own and one enemy pawn
- on e file you have just one enemy pawn, but no own pawn

c5 is of course due bigger penalty

[d]6k1/8/8/3p4/3P4/3P4/8/6K1 w - - 0 1
symmetrical d4 doubled pawn - both adjacent c and f files are open, no own or enemy pawns on them

Symmetrical doubled opposed pawns are due bigger penalty, maybe some 10-15cps, more in the endgame.

The requirement that the pawn is opposed is essential, as otherwise the doubled pawn itself breaks the symmetry. Such a pawn would not be more difficult to undouble, even if the symmetry on the 2 adjacent files is full, besides it leaves a semiopen file for the enemy.

[d]6k1/5p2/2p1p3/8/3PP3/2PP4/8/6K1 w - - 0 1
doubled unopposed pawn on d4; although there is a symmetry in terms of pawn placements on the 2 adjacent files, one own and one enemy pawn each, this doubled pawn is not due bigger penalty, as it is unopposed.

Indeed, you need a complete symmetry for the 3 files, the files where the doubled pawn is as well as the 2 adjacent files, in order for the rule to hold and higher penalty be applied.

So, again, you give higher penalty when:

- the doubled pawn is opposed
- the 2 adjacent files to the doubled pawn are either both closed, either both open, or one is open and the other one closed

I think this is a very sound rule for doubled pawns indeed, but have no clue if this will give something in engines - obviously this will depend on the particular engine and what specific eval terms it has and has not.

[Lyudmil Tsvetkov]

Well, I do not know what a vertex is, but I know even less what a vortex is, so I decided to call it like that.

A vertex is a pawn that is defended by 2 pawns at the same time, each one of which is in turn defended by another own pawn, so that a triangle of 2 separate long chains of 3 pawns each is formed with a common point/vertex the evaluated twice defended pawn.

[d]6k1/8/8/3P4/2P1P3/1P3P2/8/6K1 w - - 0 1
d5 is a vertex/vertex pawn of the 2 white long chains - it is defended by both c4 and e4, and c4 and e4 are in turn defended to the left and to the right respectively

[d]6k1/8/5P2/4P1P1/3P3P/8/8/6K1 w - - 0 1
vertex pawn on f6

Vertices are due some 10-15cps additional bonus.

Is not it enough to score just the defended pawns and the long chains?

Well, no, scoring the 2 long chains scores them separately, but the existence of a vertex means those chains are connected into a single whole, and a single whole of 2 separate chains is always worth more than its constituent parts.

Besides, such vertices are usually very mobile in moving forward, leaving behind a flexible pawn structure, hence the bonus.

The existence of a twice defended pawn, even if scored, changes nothing - this is an overbonus.