SjaakII 1.0 RC1

[hgm]

Well, never mind the locusts. On second thought it seems rather pointless to capture as a Checker when you can already take the same piece in replacement capture.

About piece-value guestimates:

Funny enough I never found a significant difference between the value of moves in the first (3x3) and second (5x5) ring. Knight is about worth as much as Commoner. It seems that what the Commoner has in terms of 'concentration' (of which its mating potential is a consequence, but which makes it also very effective against Pawns), the Knight makes up for it terms of 'speed'. These are global properties of the move pattern as a whole, rather than a sum over contributions of individual moves. For non-captures 'speed' seems to be the important aspect, for captures 'concentration'. (The piece that moved as Knight and captured as King was about 50cP more valuable than any other divergent combination of those moves.)

Range-3 leaps are considered very dangerous, but this might for a large part be the consequence of the FIDE setup, with the pieces initially totally immobilized by the Pawn wall.

Also remarkable was that the value of a Pawn does seem to hardly depend on the value of the piece it promotes to (as long as this is not very weak). I once tried to balance Spartan Chess by restricting promotion to Captain there, figuring that 8 times the value drop of the Hoplit caused by this would have a significant impact on the strength of the Spartan army. But I measured none at all! This could be partly due to elephantiasis pulling down the FIDE Pawn value, (you cannot avoid trading the Pawns for Hoplits), but my principal explanation is that it really does not matter much to what you promote, as long as it is worth at least as much as the opponent's weakest non-Pawn. Because that is what he is going to sac to prevent your promotion, or what will be tied up in stopping your passer. So a Pawn that promotes to Knight in a game where the Knight is the weakest piece is worth more than a Pawn that promotes to Queen in a game where Ferzes and Wazirs are the weakest pieces!

[Evert]

Funny enough I never found a significant difference between the value of moves in the first (3x3) and second (5x5) ring. Knight is about worth as much as Commoner. It seems that what the Commoner has in terms of 'concentration' (of which its mating potential is a consequence, but which makes it also very effective against Pawns), the Knight makes up for it terms of 'speed'. These are global properties of the move pattern as a whole, rather than a sum over contributions of individual moves. For non-captures 'speed' seems to be the important aspect, for captures 'concentration'. (The piece that moved as Knight and captured as King was about 50cP more valuable than any other divergent combination of those moves.)

Interesting.
I think my first attempt was just based on 3x3, but that's no good because you can't tell a Commoner from a Queen (and a Knight has no moves at all). The main effect (in terms of chess pieces) of including both is setting the rook/knight value ratio.

Range-3 leaps are considered very dangerous, but this might for a large part be the consequence of the FIDE setup, with the pieces initially totally immobilized by the Pawn wall.

To be honest, I never thought much about range-3 leapers. That they get dropped from the estimate is an oversight... I should do something about that.

Also remarkable was that the value of a Pawn does seem to hardly depend on the value of the piece it promotes to (as long as this is not very weak). I once tried to balance Spartan Chess by restricting promotion to Captain there, figuring that 8 times the value drop of the Hoplit caused by this would have a significant impact on the strength of the Spartan army. But I measured none at all!

The funny thing about the Spartan Captain is that it has a 5x5 attack pattern that is the same as a Rook (and in fact it is more dangerous because it cannot be blocked). I'm not very shocked that restricting promotion to Captain doesn't make much of a difference since it's not that weak.

This could be partly due to elephantiasis pulling down the FIDE Pawn value, (you cannot avoid trading the Pawns for Hoplits), but my principal explanation is that it really does not matter much to what you promote, as long as it is worth at least as much as the opponent's weakest non-Pawn. Because that is what he is going to sac to prevent your promotion, or what will be tied up in stopping your passer.

That's basically elephantiasis due to the promoted piece, right? I think that is probably a very important term here.

So a Pawn that promotes to Knight in a game where the Knight is the weakest piece is worth more than a Pawn that promotes to Queen in a game where the Ferzes and Wazirs are the weakest pieces!

Isn't that basically elephantiasis of the pawn due to the presence of the Ferzes and Wazirs? It's not quite the same, since the pawn is still worth less than either, but clearly the value of a passer approaches a Knight in the first example, and a Ferz in the second.

[hgm]

The funny thing about the Spartan Captain is that it has a 5x5 attack pattern that is the same as a Rook (and in fact it is more dangerous because it cannot be blocked). I'm not very shocked that restricting promotion to Captain doesn't make much of a difference since it's not that weak.

It didn't test any stronger than Knight.

That's basically elephantiasis due to the promoted piece, right? I think that is probably a very important term here.

No, it is elephantiasis because of the Hoplites. You can only have an advantage from Pawns being stronger than Hoplites when you avoid trading them for Hoplites. But that avoidance makes them less valuable, and the Hoplites aren't suffering a similar depression.

Isn't that basically elephantiasis of the pawn due to the presence of the Ferzes and Wazirs? It's not quite the same, since the pawn is still worth less than either, but clearly the value of a passer approaches a Knight in the first example, and a Ferz in the second.

Well, it is not really elephantiasis, as this is a tactical issue, the piece getting tactically less useful because you cannot afford it to be traded. It is more a problem with pieces that are valuable for other reasons than their tactical ability. Although you are right that in a sense it is similar, if you consider promotion tactics.

[Evert]

I am not sure pushing Pawns very aggressively is good strategy in Shogi. Pushing them should have a positive score, however, otherwise it starts po play very indecisively.

I'll have to do the experiment to make sure. As it is, I think the term that makes it push pawns is very small; it relies on advancing the game-phase to really start pushing pawns.

I ran a simple test overnight that just awards a rank-based bonus to the pawn PSQ in the middle game, and that proved to be a very nice Elo gain very quickly. There's probably more to be had here.

[Evert]

The funny thing about the Spartan Captain is that it has a 5x5 attack pattern that is the same as a Rook (and in fact it is more dangerous because it cannot be blocked). I'm not very shocked that restricting promotion to Captain doesn't make much of a difference since it's not that weak.

It didn't test any stronger than Knight.

I know, which is something I've never fully understood because it seems like it "should" be stronger. In so far that there is anything to understand when it comes to piece values.

Note though that in the first part I'm talking about a 5x5 board; I suspect a Captain may be stronger than a Knight there, but perhaps not on a 8x8 board. They also have a somewhat different movement pattern in the sense that a Captain can very quickly reach squares that are close to its current position, which is more difficult for a knight (in particular orthogonal adjacent squares). When it comes to promotion, that is usually more important in the end game and I suspect the Captain really is stronger than a Knight in the end game (it also has mate potential, which is probably utterly irrelevant with respect to its piece value).

The flip side of all that is that the scoring was based on how dangerous a piece is in a king-attack. Knights can be quite dangerous there as well (which perhaps isn't reflected very well in the score that comes out at the moment).

[myfish]

For pieces that demote on capture (in Shogi/Bughouse) a further correction is applied. If the piece has the same move as an unpromoted piece (think Tokin versus Gold), then it gets a bonus to its piece value that is larger if the piece it demotes to is less valuable. This makes a Tokin worth more than a Gold. I'm not entirely happy with how this term comes out, quantitatively.

Would you know by how much ?
If we take lance, knight and silver too, they should be worth proportionally 'more' than 'gold'.

[hgm]

It didn't test any stronger than Knight.

I know, which is something I've never fully understood because it seems like it "should" be stronger. In so far that there is anything to understand when it comes to piece values.

Why do you think it should? They both have 8 unblockable moves. And the Knight is faster: it can make a (6,6) journey in 4 moves, the Captain would need 6 moves for that. Speed is important to stop passers.

On an infinite board the number of squares that could be reached by a Knight vs Captain within 1, 2, 3... moves is: 8 vs 8, 40 vs 32, 108 vs 72, 204 vs 128. I think this should count for something. It of course means that the Captain will have on average more different paths to the same square, but I guess the advantage of that quickly saturates.

The flip side of all that is that the scoring was based on how dangerous a piece is in a king-attack. Knights can be quite dangerous there as well (which perhaps isn't reflected very well in the score that comes out at the moment).

It is true that distant moves are more dangerous to a King, and in particular distant jumps. I never really measured how much a leap should be penalized for being lame. I did measure XQ Horse, where it lost half the value compared to Knight, but this is really a pathological case, where one blocker blocks two moves. Intuitively I expect a difference for the case where the blocking square is attacked (i.e. the lame leap is really a slide) and where it can not.

Making a diagonal move to the second square a jump must be worth a lot, as the FA is worth about the same as a lone Bishop, so the privilege to jump is almost worth the same as all d>2 moves together! (Not so for Rook, though.) This could be interpreted as evidence that jumps to the second ring are actually worth more than steps to the first ring.

There could be a bonus for orthogonally adjacent squares in the move pattern. (Perhaps because this is how Pawns move?) This would explain why the Rook benefits so much more from its distant moves than the Bishop. It would also explain why there is more synergy between N and B than between N and R. (But not why there is more synergy between N and B than between R and B.)

[Evert]

For pieces that demote on capture (in Shogi/Bughouse) a further correction is applied. If the piece has the same move as an unpromoted piece (think Tokin versus Gold), then it gets a bonus to its piece value that is larger if the piece it demotes to is less valuable. This makes a Tokin worth more than a Gold. I'm not entirely happy with how this term comes out, quantitatively.

Would you know by how much ?
If we take lance, knight and silver too, they should be worth proportionally 'more' than 'gold'.

SjaakII's heuristic gives the following piece values for Shogi:

Code: Select all

P:   16
G:  111
+P: 138
+L: 131
+N: 128
+S: 113

Take with a healthy helping of salt though! It gets the ordering right, but it was designed to do so. Note too that these are not the values that Sjaak uses when it actually plays Shogi (because the piece values are given as input).

[myfish]

For pieces that demote on capture (in Shogi/Bughouse) a further correction is applied. If the piece has the same move as an unpromoted piece (think Tokin versus Gold), then it gets a bonus to its piece value that is larger if the piece it demotes to is less valuable. This makes a Tokin worth more than a Gold. I'm not entirely happy with how this term comes out, quantitatively.

Would you know by how much ?
If we take lance, knight and silver too, they should be worth proportionally 'more' than 'gold'.

SjaakII's heuristic gives the following piece values for Shogi:

Code: Select all

P:   16
G:  111
+P: 138
+L: 131
+N: 128
+S: 113

Take with a healthy helping of salt though! It gets the ordering right, but it was designed to do so. Note too that these are not the values that Sjaak uses when it actually plays Shogi (because the piece values are given as input).

Yes it does indeed order them correctly.

[Evert]

I know, which is something I've never fully understood because it seems like it "should" be stronger. In so far that there is anything to understand when it comes to piece values.

Why do you think it should? They both have 8 unblockable moves. And the Knight is faster: it can make a (6,6) journey in 4 moves, the Captain would need 6 moves for that. Speed is important to stop passers.

I guess this is a question of what makes a piece more valuable. My thought that it "should" be stronger is based on king-safety rather than the ability to stop passers. As you say, Knights are faster if they need to move a great distance that is not along a straight line. I guess you could say they are more "agile".

The heuristic seems to be quite bad for Spartan minors: it puts the Captain at 500 (same as a rook, unsurprising considering what it measures) and the Lieutenant at a whopping 666 (Knight/Bishop are 333).

Making a diagonal move to the second square a jump must be worth a lot, as the FA is worth about the same as a lone Bishop, so the privilege to jump is almost worth the same as all d>2 moves together! (Not so for Rook, though.) This could be interpreted as evidence that jumps to the second ring are actually worth more than steps to the first ring.

There could be a bonus for orthogonally adjacent squares in the move pattern. (Perhaps because this is how Pawns move?) This would explain why the Rook benefits so much more from its distant moves than the Bishop. It would also explain why there is more synergy between N and B than between N and R. (But not why there is more synergy between N and B than between R and B.)

How pawns move is probably a very important term. It's also very complicated.

The piece-value model used by Sjaak is extremely crude and simple and based on only one metric: king-safety (this is because the king-safety metric is something I already had and I want to implement something quickly). This may be somewhat adequate if king-safety is what you're interested in (say, for MG values), but for the end-game it becomes important whether you need to worry about passers at all (in Sittuyin you don't, in Makruk you need to worry a lot less than you do in Chess) and then whether you can intercept them easily or not, which depends on how pawns move. It would be interesting to do piece-value measurements for Legan's Chess (where pawns are lfmFlfcW). Now it's Bishops that can support passed pawns and Rooks that need to worry about slipping through openings in the pawn structure. If passers are not a concern, whether a piece can deliver mate or not may (finally!) become a concern for piece value.