Bored with orthodox Chess?

[Richard Allbert]

I can send it you... can you pm me your email? Or alternatively I'm richard dot allbert at gmail dot com

I can't seem to access JabbaChess.blogspot.com, which is a bit odd. Seems to have been a problem Google merged two accounts I had.

Also Fileave.com have suspended my account because I haven't logged on for 30 days...

I'll look into to it tomorrow.

Ciao

Richard

[Richard Allbert]

It's on

[url]ttp://www.filehosting.org/file/details/193181/Catalyst.zip[/url]

No hash tables, and it loses 20 from 30 games vs Fairy... I need to do a bit of tweaking!

Richard

[hgm]

OK, got it. I will try it against ChessV then.

One request: can you add a feature

feature variants="spartan"

in a future version? That is not really needed to run it when you start WB already in variant spartan, but it makes it possible to switch to Spartan from the WinBoard menu without drawing a complaint from WinBoard.

[Richard Allbert]

Done, it's here..

http://www.filehosting.org/file/details ... talyst.zip

[George Tsavdaris]

You have to see it in action to judge it. Perhaps I should set up a server running demo games. Spartan Chess has many interesting unusual features. But the main reason I am so impressed with it, is that it does not seem to have any flaws. The distribution of piece values is such that you get very natural evolution of the game towards an end-game,

Well yes and exactly this is a great flaw!
I've seen you say that you perhaps lean to believe that Spartan Chess may be even better than Chess.
Now i've seen the game: No way!

Yes it's well balanced, yes all the new concepts are great and imaginative, yes endgames have a certain type of dynamics with no draw on the horizon as it mostly happens in Chess, BUT, a big but is that it misses the magic of Chess. What is this?
Attacking combinations and sacrifices or exchange sacrifices!

It's far far more difficult to create an attack to the King in Spartan Chess. No more sacrifices. No more Rook for Knight or Knight/Bishop for Pawn sacrifices. The short range of the Spartan army forbids that. Spartan army can't afford to give material in the middlegame.

All in all is a great game but it is way behind Chess.


BTW i have realized these days that i'm playing this game, that there is a flaw in Hoplites and Pawns strength distribution.
Pawns can attack with 2 ways and can move with 1. Hoplites can move with 2 ways and can attack with 1.
So you can block Pawns with 1 way only, while you can block Hoplites with 2.
Double reason.

It's a clear disadvantage of Hoplites movement compared to Pawns in opening and middlegame.
In endgame since there are not much pieces remaining, the blocking part is eliminated and the movement in 2 directions may even help them! But Spartans have first to survive reaching an endgame.

BTW what values you have put to all 12 pieces of Spartan Chess? I'm speaking about Fairy-Max but since it's relatively primitive and you may have used heavy approximations, what values you would have put to Joker for example?
I'm very interested in the Spartan values(as also in Persian values whether you have kept the same as in Chess), as also to the 2 Kings of Spartan although setting the values to Kings is more tricky.


And if you really didn't want to rely on your personal opinion setting those values, what (quick-not taking weeks) method you would have used to derive them?
I'm aware of "Safe Check" method and Reinhard Scharnagl's method but they don't seem to work for leapers like Captain or Lieutenant.
You can still, for safe check method, make the assumption to the calculations to include positions where a piece is standing in front of the King so the leaper safe checks him, but then the calculations are very hard. (Or not? Hm.... I will try. )

But do you know of any method other than these 2 for deriving good pieces values?

[hgm]

Thank you for the analysis of this variant from a human perspective. As to the piece values:

I have not found an 'ab initio' way to calclate slider values yet. So I usually rely on empirical determination of piece values: just make a substitution in the FIDE setup, e.g. Captains for Knights for only one side, play 400-1000 bullet games, and see if the side with the Knights or with the Captains does better. And then try to compensate a significant advantage by Pawn odds, and an overwhelming advantage by making another substitution. This, however, is a tedious process.

From making such systematic measurements, I learned that neither Reinhard's nor Ed's method is any good. They cannot explain why The difference Queen vs Archbishop is smaller than a Pawn, and Chancellor minus Archbishop much less than half a Pawn (A+A+P significantly beats C+C). The A values they predict are typically 2 Pawns too low. I can of course also not explain that; everything I could think of (mating potential, color binding, forwardness) fails misarably when applied to other pieces. So it is just an observation, but a very well tested one (in many different piece combinations).

For short-range leapers (jumping 2 squares maximum) I found an approximate law for their piece values. These mainly depend on the number of squares N they can move to, as

Value (cP) = (30 + 5/8*N)*N

This would give:

4 (Wazir, Ferz): 130
8 (Knight, Captain, Commoner): 280
12: 450
16: 640
24: 1080

For divergent or asymetric pieces there is the ruleof thumb that captures contribute twice as much as noncaptures, and that forward moves contribute twce as much as sideway or backward moves. (Determined by disabling individual moves on pieces wth many moves.)

Within this approximation there is no difference between symmetric pieces with the same number of moves, while in practice there is. This is to be expected: pieces have 'global' properties that depend on how well their moves work together to acheive certain goals. Like mating potential, speed, color binding, focusing. Mating potential seems not to count for much (with the strong caveat that this has been measured with a stupid engine that would not realize you can draw by sacrificing your piece for the last Pawns if the opponent only has a minor). Speed is more important (to catch and stop passers). Hence a Knight is stronger than a Commoner. Actually a piece that captures like a King and non-captureslike a Knight is half a Pawn stronger than any other combination (i.e. capture as Knight, move as King, or plain Knight or King). Color binding is not a big deal if you have a pair of the pieces, but it makes part of the value come from a pair bonus. (Also here a caveat, as the engine did not implement pair bonuses.)

With my current understanding, I would program the following piece values in an advanced engine:

P = 80 (but all kind of cooperative bonuses)
N = 325
B = 325 (+ 50 for pair)
R = 475 (+ 50 for open file)
Q = 950

H = 75 (?)
C =300
L = 350
G = 725
W =875
K = 450 (but heavily dependent on safety of the other King)

Note that the value of the spare King is much larger than that of a Commoner, which would only be 275-300. The piece values for the Persians are the standard values for FIDE Chess.

[Uri Blass]

For divergent or asymetric pieces there is the ruleof thumb that captures contribute twice as much as noncaptures

If I understand correctly
based on this rule a piece that can capture like a queen but cannot move
has 2/3 of the value of the queen and a piece that can move like a queen but cannot capture has 1/3 of the value of the queen(and the same for a rook a bishop or a knight).

You may be correct but I wonder if this opinion is tested in games.

Note that if you want to test chess engine with different rules then then simplest rule to implement is to have pieces that are different from normal chess pieces by the fact that they cannot move or the fact that they cannot capture.

Edit:Thinking about it again you need to substract the value of a piece that cannot move and cannot capture in case that it has some positive or negative value.

I am not sure if it is better to have some white stone that cannot move in some square relative to not having it.

Not that the stone is not symmetric and when the opponent can capture it the side that has the stone cannot do it and it may block the pieces of that side.

[hgm]

I have not tested pieces that cannot move or cannot capture at all. Like I wrote, my statement was about pieces that had many moves, where I determined the marginal effect of disabling one of the captures or non-captures. It stands to reason that the value of remaining captures would suffer from a lage decrease of the number of moves, as you would have more difficulty aiming the captures at something. In the extreme case of having no non-captures at all, the opponent can simply avoid your captures. Similarly, if a piece has no captures, you can move it all you want, but it will not be very dangerous wherever it is. So moving it does not do you much good.

In fact the (30+5/8*N)*N 'law' already shows cooperativity between moves, rather than simple additivity, through the quadratic term. A generalization to divergent pieces would be (1+a*C+b*NC)*(c*C+d*NC), where C = nr of captures, and NC = nr of non-captures. My suspicion is that in this case d<<c, (i.e. it is mostly the captures that make a piece valuable).

What I did test in games was the value of a piece that captured as Q and non-captured as N (mNcQ), which had value ~7, and mQcN, which has value ~5. So the ratio 1:2 seems to hold reasonably there. But, like I already mentioned, there can be surprises: mNcK is stronger than both N and (non-royal) K. But there both components had N=8, so it is the more subtle effects that tip the balance there.

[PK]

a little debugging question: which letters are expected to denote promoted pieces? I have tried to run a couple of games against micro-max and f7e8a (like archbishop) has been rejected.

[hgm]

The piece names in Spartan Chess are:

H = Hoplite
C = Captain, jumps 1 or 2 orthogonal
L = Lieutenant, jumps 1 or 2 diagonal (+ sideway non-capture)
G = General, moves as K+R
W = Warlord, moves as B+N
K = King

In principle black can promote to any of the Spartan pieces, even to King if he only has one, but I suspect this is never useful.

So to promote to "archbishop" you would need something like e2d1w.