Shokidoki wins silver at ICGA Olympiad for 5x5 Shogi

[hgm]

My Shogi engine Shokidoki turns out to be much stronger in mini-Shogi(with 6 pieces per side, on a 5x5 board) than in regular Shogi (on 9x9). I had not planned to even participate in 5x5 Shogi, because I did not know it existed. But when I learned (after coming here) that it is basically ordinary Shogi on a 5x5 board (with the same, but fewer pieces), it was trivial to convert Shokidoki to play that variant too. So over the weekend I spent my nights in the hotel to adapt the board size. And because I had nothing else to do on Tuesday (Chinese Chess already finished, and regular Shogi starts tomorrow), I asked if they would still allow me to participate. And because they had an odd number of participants, this was no problem.

To everyone's surprise, (and most of all my own), the algorithm used in Shokidoki (with a capture search for Quiescence as in Chess), which works only so-so in regular Shogi, seems perfect for mini-Shogi. Of course I was clobbered in the first few rounds, because my piece values were all wrong for the 5x5 board (a Rook has at most 8 moves there, and is thus hardly worth more than a Gold General), and my King Safety allowed the King to stray way too much forward (it turns out to be really essential to stay on the back-rank, and preferably in the corner in mini-Shogi).

But I was able to fix those apparent defects during the lunch break, by changing the piece values / piece-square tables. I must have had a very lucky hand in picking the new values, (there was no time for any testing, of course), because after that Shokdoki ran like a charm, and won all its games! So it came in second from 8 participants, with 11 wins and 3 losses:

1. (14 pt) Clairvoyant 1/128
2. (11 pt) Shokidoki 0.8m
3. (10 pt) 55Tacos

[Graham Banks]

Congratulations H.G.

[Jack Lad]

Well done Prof Muller!

I am surprised that such a tournament exists because I thought that this simple game would always be easily drawn (via the 3 move repetition rule) between decent engines.
http://en.wikipedia.org/wiki/Minishogi

Perhaps you could make it play "Let's Catch the Lion!" aswell. This game is probably easily drawn but I would like to know if this is the case.
http://en.wikipedia.org/wiki/Dobutsu_shogi

[hgm]

They actually adopted the rule here that he who repeats loses (unless the other side is perpetually checking; then that side loses).

[shogi4fun]

Also congratulations from me.

Chinese Chess also finished?
Where can I read the results?

I expect here, but still not updated, yet.

Oh, and of course good luck for your Shogi tournament.

[Gerd Isenberg]

Wow! Congrats as well.

Cheers,
Gerd

[beachknight]

my congrats as well.

best,

[gerold]

Nice going H.G.

Good luck,
Gerold.

[bob]

My Shogi engine Shokidoki turns out to be much stronger in mini-Shogi(with 6 pieces per side, on a 5x5 board) than in regular Shogi (on 9x9). I had not planned to even participate in 5x5 Shogi, because I did not know it existed. But when I learned (after coming here) that it is basically ordinary Shogi on a 5x5 board (with the same, but fewer pieces), it was trivial to convert Shokidoki to play that variant too. So over the weekend I spent my nights in the hotel to adapt the board size. And because I had nothing else to do on Tuesday (Chinese Chess already finished, and regular Shogi starts tomorrow), I asked if they would still allow me to participate. And because they had an odd number of participants, this was no problem.

To everyone's surprise, (and most of all my own), the algorithm used in Shokidoki (with a capture search for Quiescence as in Chess), which works only so-so in regular Shogi, seems perfect for mini-Shogi. Of course I was clobbered in the first few rounds, because my piece values were all wrong for the 5x5 board (a Rook has at most 8 moves there, and is thus hardly worth more than a Gold General), and my King Safety allowed the King to stray way too much forward (it turns out to be really essential to stay on the back-rank, and preferably in the corner in mini-Shogi).

But I was able to fix those apparent defects during the lunch break, by changing the piece values / piece-square tables. I must have had a very lucky hand in picking the new values, (there was no time for any testing, of course), because after that Shokdoki ran like a charm, and won all its games! So it came in second from 8 participants, with 11 wins and 3 losses:

1. (14 pt) Clairvoyant 1/128
2. (11 pt) Shokidoki 0.8m
3. (10 pt) 55Tacos

Congrats...

I finished in 4th place at the 1976 ACM computer chess tournament. During that event, I rewrote major parts of that program when I suddenly discovered a glaring software problem that was major in its impact. It worked.

I have seen programmers tweaking between each round, many times since then. I have gotten out of that mode, trying to do all my testing _before_ an event, rather than during, as that avoids introducing unexpected bugs that cause ugly play.

[uaf]

Congratulations HG.

This must be one of there rare case where the changes made during a tournament improve the program rather than introducing new bugs