Ongoing GO programming?

[rjgibert]

To be frank, the interest that this chess variant garners is quite surprising. The first impression it gives is negative and unfortunately, it is not the type of game where a players appreciation of it grows over time.

A few years ago, I used to play it on ICC and managed to top the rankings among the human players. It sounds like I must have liked it, but actually, what attracted me was not the game itself, but rather the idea that I could manage to be number one in the world ICC-wise at one of the game variants offered there. I found this quite amusing, since I honestly could not have been much of a good player. My interest in shatranj began and ended at that.

The game itself is quite poorly designed. In short, it's crap. I stop short of calling it pure crap, because it did after all inspire the game of modern chess. The genius the Italians imparted to this end cannot to be underestimated. For example, in endgame studies, with a small number of pieces, a result of tremendous beauty and ingenuity is possible in the modern game. In this one respect, chess contrasts quite sharply with a game like go.

BTW, I feel compelled to confess that the high rating I got was mostly due to the fact that the vast majority of the games I played were against a single computer opponent called elephant, which frankly played terrible. This in combination with the carry over of the skills I acquired in regular chess where I am rated over 2200 USCF made the number one ranking possible, so there was a significant amount of a rating inflation effect that I was exploiting. I did play other people and programs, but this represented a small percentage of the games I played.

Currently, the best player on ICC appears to be the computer account Oliphaunt. This is a crafty clone specialized to play shatranj. ICC gives the email of its programmer as OrionsKnight@chessclub.com. You may want to check this program out.

[yoshiharu]

To be frank, the interest that this chess variant garners is quite surprising.

Actually I find so hard to find infos and stuff about shatranj that I didn't think there was _any_ interest at all...
You find stuff on go, shogi (many variants), backgammon, checkers...even nim, but not much about shatranj

My interest in Shatranj is actually motivated by
a) history of Chess
b) I started programming a an engine playing a pawn-only game I myself developed (quite a stupid one, though), then I expanded it to play Los Alamos Chess, then to Shatranj, and now I'm in Int'l Chess; a hierarchy of toy models, to prepare the conquer of the world

Currently, the best player on ICC appears to be the computer account Oliphaunt. This is a crafty clone specialized to play shatranj. ICC gives the email of its programmer as OrionsKnight@chessclub.com. You may want to check this program out.

Thanks for the info!
I will check this out.
Cheers, Mauro

[yoshiharu]

Fairy-Max is a Winboard engine that can be configured for Shatranj, although I am not sure about the repetition and reversible-move draw rules in that game (or even how stalemate is handled). So it might handle those completely wrong. Also the piece values are wild guesses. But it should play, even if lousy.

Yesterday I tried Fairy-Max; actually works fine, except it seems to not handle correctly the "edit" command, so that I cannot play a Nunn(-like) match; also, since AFAIK in Shatranj there is no castling, I removed it from the suggested .ini: I will do some guesswork on the values of the pieces, anyways.
One thing went wrong, instead:
in this position (B and Q stand for Alfil and Ferz)

6nr/R1qk4/3b2pp/3p4/2P2p2/1N3K1B/2P2PPP/1rBQR3 b - - 0 30

Fairy-Max, as the Black, played
30...,d4c3

I don't know where is the problem, unfortunately I don't have the debug log of winboard.

Anyway I have to thank you, since this event triggered, later in the game, a bug I introduced when I ported the check evasion generator from my Chess engine to the Shatranj one

Regarding the victory-loss conditions: have you thought of codifying them into the .ini file (in some format)?

Cheers, Mauro

[hgm]

Hmm, it plays d4c3 here:



Without debug we will not be able to trace it, though.

You are right about the castling; I will remove it from the fmax.ini.

As for the edit command: with me it seems to work perfectly. But with the fmax.ini as it is given you have to use E (Elephant) for Alfil, otherwise you get an ordinary Bishop. (Both F and Q give you a Ferz, though. I had to do that, as micro-Max always promotes to Queen. So I had to redefine the Queen as Ferz, and to make sure that fMax understands the F in the edit menu I defined a second Ferz in the F position.)

If you want B to mean Alfil, as your FEN above suggests, you should replace the line for B by that for E. Otherwise B will remain an ordinary Bishop.

[yoshiharu]

As for the edit command: with me it seems to work perfectly. But with the fmax.ini as it is given you have to use E (Elephant) for Alfil, otherwise you get an ordinary Bishop. (Both F and Q give you a Ferz, though. I had to do that, as micro-Max always promotes to Queen. So I had to redefine the Queen as Ferz, and to make sure that fMax understands the F in the edit menu I defined a second Ferz in the F position.)

If you want B to mean Alfil, as your FEN above suggests, you should replace the line for B by that for E. Otherwise B will remain an ordinary Bishop.

I forgot mentioning: I didn't download the executable, I compiled it from the source (using Dev-C++, basically a gcc windows port). Moreover I am using winboard_x with legality test switched off. Can these things influence the engine?
I tried to swap the E and B lines, but it didn't work out: problably I didn't understand exactly how the .ini file works...

Cheers, Mauro

[hgm]

Windows_x probably doesn't accept FENs with an E and F in it. Fairy-Max uses protoversion 1, which means the position is loaded with the edit command, and Winboard_x will be using Q and B there, as in the FEN.

In principle, this should do it:

Code: Select all

8,8
6 4 5 7 3 5 4 6
6 4 5 7 3 5 4 6
P:3 -16,6 -15,5 -17,5 
p:3 16,6 15,5 17,5
K:-1 1,7 16,7 15,7 17,7 -1,7 -16,7 -15,7 -17,7
N:15 14,7 31,7 33,7 18,7 -14,7 -31,7 -33,7 -18,7
E:3 30,7 34,7 -30,7 -34,7
R:25 1,3 16,3 -1,3 -16,3
F:8 15,7 17,7 -15,7 -17,7

The first 3 lines are board size and white and black initial piece setup.

After that come the pieces (referred to as 1, 2, 3, ... in the setup) in the order PpKNBRQ (as referred to in the FEN / edit menu).

[jshriver]

I kinda quit work/research into Chess to persue Go development. Oddly even though it's supposedly harder to code for than Chess, I'm finding it a lot easier.

If you're interested join the computer-go maling list.

-Josh

[IanO]

I have made my best estimate of the top ten 19x19 computer Go programs on Sensei's Library, a wiki dedicated to the game of Go. This estimate is hindered by the lack of participation of top programs in recent computer Go competitions. (There is currently no equivalent to the WCCC in the Go world. The well-funded World Computer Go Championships died off with the Ing Prize in 2000. Participation in the Computer Olympiad by top programs has been spotty.) My results are summarized here; source material is at the first link above.

• 1. MoGo
  2. CrazyStone
  3. KCC Igo (Silver Star)
  4. Haruka
  5. Go4++
  6. HandTalk (Goemate)
  7. GNU Go
  8. Many Faces of Go
  9. Go Intellect
  =10. Wulu, Aya, Katsunari, Go Ahead

Notes:

MoGo and CrazyStone represent the new wave of Monte Carlo-based Go programs. There was recently a short article about the technology in Scientific American.

CrazyStone is by Remi Coulom, author of The Crazy Bishop. Other than him, there has been little successful cross-pollination between computer Go and computer Chess circles.

KCC Igo from North Korea had a lock on the Gifu Challenge title from 2003-2006. Silver Star, Haruka, and HandTalk are only sold in asian countries, as far as I know.

Ian

[shiv]

Thanks for all the replies. The information provided is very useful.

[jshriver]

I didn't realize there was a World Computer Go Championship.
For a good computer Go testbed there is the CGOS server which has a lot of buzz right now.

Perhaps once a computer capable of beating a Dan level player often is around the scene will get even more press and world championships will resume. Right now most programs can be beat by amateur players.

So the future looks interesting
-Josh