Dark Chess

[George Tsavdaris]

What do you mean by "does not even load"? You get an error message? Something should happen when you click a shortcut...

I double click the shortcut and does not load anything. It happened before and i'm almost sure that it happens when inside the .ini file of the command line of the shortcut it has a wrong line that it does not recognizes.

In that case the /firstChessProgramNames={dark} is the non-recognized line and i don't understand why. running the shortcut with the INI file to have(notice the ; in front of /firstChessProgramNames):

;/firstChessProgramNames={dark}
/variant=dark
/saveSettingsOnExit=false

The shortcut runs fine.

/variant=dark is correct. I have installed the engine as

dark /fd=..\Fairy-Max /variant=dark

(I have put it in the Fairy-Max folder), and this worked for me, some versions ago. (I might have broken something since then,

I don't think something is broken, since i now tried and it works.
I'm not sure though why. I tried everything before.

but I cannot test that now, as I am running Linux to sort out the sources.) I don't think the font could have anything to do with it. The darkening should start as soon as you start playing. (Not in the opening position after "New Game", as it does not know which side you will be playing yet.) If you are not playing you should be able to use F2 to cycle through the various flipped and darkened positions.

It works as intended by a first look.
Of course i will have to test more thoroughly to see if every kind of position is covered correctly.
I will test the engine after that.

Thanks, my mistake(although i can't seem to find what was it), no need to test anything from your part.

[George Tsavdaris]

OK i tried all possible cases i could think of, en passant, castle through check etc etc and everything seems fine with the dark squares appearance.

Only one small problem:
When pasting a game(and NOT when playing games or editing a game) and then replaying it in dark mode for one side, when a piece captures another, the other piece remains in the background until the foreground piece(the one that made the capture and the "owner" of the square) moves again.
Example:





Other than that everything are smooth.

The Dark engine plays correctly also until now. The strength of course is obviously unbelievably low.

[hgm]

OK, thans for spotting this. It was a general bug in move animation;it wold even occur in variant normal when you made a capture through click-click moving. It just did not redraw the square after animation, because it figured it already put the piece on it. Which works well-enough on non-captures. Drag-drop moving does not suffer from it, because it doesnot anmate the move again if you already dragged the piece there. And highlighting with arrow also masks it, because the drawing of the arrow forces a redraw of the square. (This is why I had not noticed it; I always highlight with arrow.)

It should be fixed now.

Sources are also on-line (except for this last fix), in the hgm.nubati.net repository, branch 'alien'.

And yes, the engine is incredibly weak; it is little more than a random mover, with a bias to good or equal captures and high visibility and protected pieces, and against exposing pieces to capture by lower visible pieces, to Pawn capture from invisible squares and its King against (possible) capture in general. But it only served to test the protocol.

[Daniel Shawul]

To HG & George:

I would like to try monte-carlo tree search for a stochastic game. All games I have implemented so far are deterministic (perfect information) so I want to try out this game.
I hope it is not very hard. I will try to understand the rules of the game. If you have any pointers & suggestions, I would like to hear them.

[hgm]

I am not sure what exatly you mean by 'stochastic game', but I suspect Dark Chess doesn't fit the bill. The problem of Dark Chess is that it is a game of incomplete information, and that both players have different information.

[Daniel Shawul]

Yes I realized that is vague and edited my post accordingly. But anyway aren't imperfect information games all probablistic (stochastic I suppose?).
Alpha-beta would obviously not work there as it only searches a single variation, I thought a UCT would be more appropriate there. I guess I am wrong ...