Tuesday, March 4, 2008

Vortex vs. Joker80, a Wild Game!



Last night Gothic Vortex played an hour-long game against H.G. Muller's Joker80 program. I must say, this was the most unusual game of Gothic Chess I have ever seen! Shown above is the position after 16. Ae3!? to this game:

Vortex vs. Joker80

I will annotate this game later, but feel free to comment on it here.

6 comments:

H.G.Muller said...

This game is way beyond my abilities to comment on it. I am not that good a Chess player. What I notice is that Vortex seems to have hanging pieces all the time, but Joker80 refuses to take them. Apparently because it thinks it can gain even more. Without having the PGN file where Joker stored its own scores, I cannot even guess what it was trying to accomplish, and where it miscalculated. It might have overvalued its own advanced passers, and overlooked that the a-Pawn finally could be stopped by the Chancellor. After that, a R+N vs C end-game is left, which is definitely not good for Joker. (Although Vortex' tripled Pawns probably hardly count as more than one.)

Again, the problem here might be that (contrary to what my piece-value measurements indicate) Joker values Knights about 50 cP too high. It should have used the Knight to defend against the advancing connected passers, but it was just sitting there as a dummy until it was too late.

GothicChessInventor said...

The hanging pieces situation occurs because of some very complex and long-range tactical combinations that present themself. When two program of high caliber are facing one another, the positions that come about are usually the result of the "granular holes" in their evaluation functions. Joker80 is going to get Gothic Vortex into positions where Vortex underestimates the opponent's chances and perhaps Joker80 over-evaluates its own chances, and Gothic Vortex is going to get Joker80 into positions where Joker80 underestimates Vortex's chances and/or Vortex has over-estimated its chances.

So, what you see is a tug-of-war as each search tries to put the other side into even more peril.

You might have noticed the tripled pawns were present for a good part of the game. You should really only "discount" them once they are off the board :)

It was a very complex game, and when I get somet time I will annotate it for the blog.

H.G.Muller said...

Well, to judge winning probabilities in a Pawn ending the rule of thumb to count doubled isolated pawns as only a single one usually works pretty well. E.g. white: Pb2,b3,b4, Kd4; black Pb5, Kd6 is a dead draw, despite the two-pawn advantage for white.

Crocodile said...

This is a very dangerous generalization.
For example if it is black to move or the white King is on another appropriate square it would be a white win. And there are countless such appropriate squares.

Or if we have for example white: Pb2,b3,b5 Ke3; black: Pb6,Kf7 then it would be a white win no matter which moves it is.

Treating double isolated or not Pawns, as one, is dangerous.....

H.G.Muller said...

any generalization is dangerous in Chess, as the tiniest detail of a position might make a huge difference. This is why it is such an interesting game.

But as a zero-order approximation, the rule doubled = 1 is not such a bad one, if you would do statistics on the number of time it holds versus the number of times it doesn't. That there are lost positions of 3 vs 1 in itself doesn't prove that the rule is bad: there are plenty of positions where 1 vs 1 is lost too.

But you are correct in pointing out that even a worthless Pawn is not completely worthless, as is might provide you a 'zugzwang bouncer', which in Pawn endings can be a winning advantage. So perhaps the rule doubled = 1 + zugzwangAdvantage would be an even better approximate rule.

GothicChessInventor said...

There is knowledge, and there is search.

At some point in time, you have to stop searching and statically evaluate a leaf node.

If a pawn is to come off the board, I let the search find it. I try not to devalue material on the board.

I do examine pawn structure that surrounds a king, and depending on the material distribution on the board, I give a bonus or inflict a penalty.

In the grand scheme of things, there are so many other factors in Gothic Chess that could adversely affect a position... Like M_TAL getting his pieces in Vortex's face and running it off the board!

:)