Sunday, November 18, 2007

fmax4 replaces Martillo in Gothic Chess Computer Championship

Due to technical difficulties, fmax4 has replaced the Martillo program in the 2007 Gothic Chess Computer World Championship. The Martillo programmer cites an example we have all seen at one time or another: making too many changes to the only copy of the source code, then messing something up in a drastic fashion.

While Martillo has stepped down, fmax4, a "famous descendant" of the smallest chess program on the planet (named "Max" by H.G. Muller of The Netherlands) has moved in to take its place. Max is a chess program with fewer than 2000 characters in its source code, yet it plays respectable chess at the "expert" level of about 2000 elo. At first, H.G. Muller was uncertain how fmax4 (the version that plays Gothic Chess) would perform.

But, with its round 1 win over Zillions 2, H.G. is pleased and will be supplying an updated version for the event.

Here is a link to that game:

fmax4 vs. Zillions 2

He notes that fmax4 is more of a "benchmarking tool" than an actual program. Because it deliberately lacks sophistication, more robust programs should be able to score wins against it. But those that cannot, clearly need to go back to the drawing board!

We welcome fmax4 into the event, which will be the most competitive it has been so far.


ChessCarpenter said...

A nice game by fmax4! There is obviously room for improvement, but an enjoyable game.
I give H.G. Mueller a lot of credit for coming up with a program to play in the 2007 Gothic Chess Computer Championship, especially with martillo pulling out with technical problems. Its great to see many people involved with Gothic Chess!

Victor Vector said...

I offer hearty congrats to H.G. for stepping up to the plate. It's tough to pass judgment on these two combatants. One was made with the least amount of code possible, the other was made to play every game that could possibly be coded! Each is at the extreme and each has the least amount of information about the game it is playing although for very different reasons.

I notice every pair of games from the first round was talked about except Variant Pulverizer vs. Chess V. Was this intentional or accidental or something else? To me it looked like each program played too much on the king's side with not a wit of concern for their own well being! To me this is the sign of the extreme amateur. Maybe early round bugs as was mentioned before by others.

H.G.Muller said...

It is true that both fMax and Zillions have little Chess-specific knowledge, but looking at the game I got the impression that fMax does have a slight advantage here. The piece values used by fMax in this game were picked by hand. Some were still off quite a bit (I only completed measurement of piece values today, and fMax will use them from the next game on), but at least they were educated guesses based on years of experience with normal Chess.

Moves like 7. e4 (after 6. e3) look a bit silly, and really are supposed to be discouraged in fMax. I discovered a bug that lead to fMax always thinking it was in the end-game, though, and it gets an overwhelming urge to push Pawns there. Next time this bug will also be corrected. (Unlike its cousin micro-Max, fMax has to count the amount of material that is on the board in the opening position as specified in its .ini game-definition file, and I left out a factor 37 there.)

Some decisive moves were really a direct consequence of the little knowledge that I did put in, though. 10. Ni5 was a conscious attempt to destroy the opponents castling rights. 18. gxf6 seemed a bad mistake by Zillions: although fMax values Bishops half a Pawn higher than Knights, and does not know about doubled Pawns, it still decides to give the Bishop, because it can force the opponent to move away the Pawn in front of a King without castling rights, which he considers worth a lot. The line that is does opened on Zillion's King later turns out decisive, and it gives me a good feeling that this was intentional.

GothicChessInventor said...

Hello H.G. and welcome aboard! Glad to hear you'll have an improved version of fmax ready for round 2. You'll have white against Zillions so you'll theoretically have a chance to really strike first in that game.

To answer victor's question: I found the ChessV game to be the poorest from among the group. Its play was weak, and Variant Pulverizer was not too impressive neither. I had a tough time finding what to say about the progams. I was told later than Variant Pulverizer had a bug, and a new version has already been submitted for Round 2. We'll have to see how it goes this time around!