Why no CrazyHouse tournament in ICGA championship ?

[Vinvin]

I see there was no a single one in all the history : http://www.game-ai-forum.org/icga-tournaments/games.php
despite some remarkable points :

1) It is quite challenging for developers : very high branching factor because many possible moves, when one could put 5 different pieces on 40 empty squares, that means over 200 possible moves.

2) Selectivity is a problem because of king safety : many deep sacrifices are possible.

3) Humans seems (still) stronger than computer at CH : Atropheide (200 Elo below top human https://fr.lichess.org/player/top/200/crazyhouse ) made some videos where he beats some engines : . But I didn't see him against the strong Imortal2.

4) There a more than 100'000 games played each month on Lichess.org

5) CH is close to Shogi, may we can see a derivative come from this world ?

[hgm]

Well, like you say the problems are similar to those of Shogi. Difference is that in Shogi you can earn big bucks by commercializing a strong engine, while no one would give a hoot about Crazyhouse.

But last ICGA Olympiad Shogi had to be cancelled, because there was only one registrant (namely me).

My Shogi engine Shokidoki plays not only the standard 9x9 game, but also many variants: mini-Shogi (5x5), Judkins Shogi (6x6), Tori Shogi (7x7), and Euro-Shogi (8x8).

I never made an attempt to also have it play Crazyhouse, although in terms of board size and number of piece types it should be able to do this. Of course the fact that there would have to be castling, e.p. capture and promotion choice would pose some technical problems, but these are only small. (I could map the Shogi promotion choice promote/defer onto the choices Queen/Knight, as Rook and Bishop seem even more useless choices in Crazyhouse than in Chess.)

I have the feeling that the strategy of Crazyhouse and Shogi iscompletely different, however. The most influential rule difference is that in Shogi you canot drop Pawns in files where there already is one.In practice this means that you can hardly ever drop Pawns. In Crazyhous dropping Pawns on 7th rank seems to be the most common way to gain material. So it would need a completely different evaluation for Pawn structure.

[bob]

Well, like you say the problems are similar to those of Shogi. Difference is that in Shogi you can earn big bucks by commercializing a strong engine, while no one would give a hoot about Crazyhouse.

But last ICGA Olympiad Shogi had to be cancelled, because there was only one registrant (namely me).

My Shogi engine Shokidoki plays not only the standard 9x9 game, but also many variants: mini-Shogi (5x5), Judkins Shogi (6x6), Tori Shogi (7x7), and Euro-Shogi (8x8).

I never made an attempt to also have it play Crazyhouse, although in terms of board size and number of piece types it should be able to do this. Of course the fact that there would have to be castling, e.p. capture and promotion choice would pose some technical problems, but these are only small. (I could map the Shogi promotion choice promote/defer onto the choices Queen/Knight, as Rook and Bishop seem even more useless choices in Crazyhouse than in Chess.)

I have the feeling that the strategy of Crazyhouse and Shogi iscompletely different, however. The most influential rule difference is that in Shogi you canot drop Pawns in files where there already is one.In practice this means that you can hardly ever drop Pawns. In Crazyhous dropping Pawns on 7th rank seems to be the most common way to gain material. So it would need a completely different evaluation for Pawn structure.

Bert Gower and I used to play CH all the time (it was mostly called "bughouse" back then. There is one serious issue, communication. Two good players can communicate with the usual "give me a knight and it's mate" and such. That adds a degree of complexity in that a good human will say "just wait, it will take me two moves to get a knight..." and the clock obviously factors in because your partner is waiting for you to give him a knight, your opponent is waiting so that you can't win the knight... etc, etc, etc...

That would be a new type of program for sure, and not just because of dropping pieces. If you think of one program playing both boards, it is also an interesting (and complicated) challenge.

I've thought about it off and on over the years, but never took the plunge...

[Vinvin]

Well, like you say the problems are similar to those of Shogi. Difference is that in Shogi you can earn big bucks by commercializing a strong engine, while no one would give a hoot about Crazyhouse.

But last ICGA Olympiad Shogi had to be cancelled, because there was only one registrant (namely me).

My Shogi engine Shokidoki plays not only the standard 9x9 game, but also many variants: mini-Shogi (5x5), Judkins Shogi (6x6), Tori Shogi (7x7), and Euro-Shogi (8x8).

I never made an attempt to also have it play Crazyhouse, although in terms of board size and number of piece types it should be able to do this. Of course the fact that there would have to be castling, e.p. capture and promotion choice would pose some technical problems, but these are only small. (I could map the Shogi promotion choice promote/defer onto the choices Queen/Knight, as Rook and Bishop seem even more useless choices in Crazyhouse than in Chess.)

I have the feeling that the strategy of Crazyhouse and Shogi iscompletely different, however. The most influential rule difference is that in Shogi you canot drop Pawns in files where there already is one.In practice this means that you can hardly ever drop Pawns. In Crazyhous dropping Pawns on 7th rank seems to be the most common way to gain material. So it would need a completely different evaluation for Pawn structure.

Bert Gower and I used to play CH all the time (it was mostly called "bughouse" back then. There is one serious issue, communication. Two good players can communicate with the usual "give me a knight and it's mate" and such. That adds a degree of complexity in that a good human will say "just wait, it will take me two moves to get a knight..." and the clock obviously factors in because your partner is waiting for you to give him a knight, your opponent is waiting so that you can't win the knight... etc, etc, etc...

That would be a new type of program for sure, and not just because of dropping pieces. If you think of one program playing both boards, it is also an interesting (and complicated) challenge.

I've thought about it off and on over the years, but never took the plunge...

Bob, you talk about "Bughouse" (2 vs 2), but the topic is the more simple (but already very tricky) Crazyhouse (1 vs 1).

[bob]

Well, like you say the problems are similar to those of Shogi. Difference is that in Shogi you can earn big bucks by commercializing a strong engine, while no one would give a hoot about Crazyhouse.

But last ICGA Olympiad Shogi had to be cancelled, because there was only one registrant (namely me).

My Shogi engine Shokidoki plays not only the standard 9x9 game, but also many variants: mini-Shogi (5x5), Judkins Shogi (6x6), Tori Shogi (7x7), and Euro-Shogi (8x8).

I never made an attempt to also have it play Crazyhouse, although in terms of board size and number of piece types it should be able to do this. Of course the fact that there would have to be castling, e.p. capture and promotion choice would pose some technical problems, but these are only small. (I could map the Shogi promotion choice promote/defer onto the choices Queen/Knight, as Rook and Bishop seem even more useless choices in Crazyhouse than in Chess.)

I have the feeling that the strategy of Crazyhouse and Shogi iscompletely different, however. The most influential rule difference is that in Shogi you canot drop Pawns in files where there already is one.In practice this means that you can hardly ever drop Pawns. In Crazyhous dropping Pawns on 7th rank seems to be the most common way to gain material. So it would need a completely different evaluation for Pawn structure.

Bert Gower and I used to play CH all the time (it was mostly called "bughouse" back then. There is one serious issue, communication. Two good players can communicate with the usual "give me a knight and it's mate" and such. That adds a degree of complexity in that a good human will say "just wait, it will take me two moves to get a knight..." and the clock obviously factors in because your partner is waiting for you to give him a knight, your opponent is waiting so that you can't win the knight... etc, etc, etc...

That would be a new type of program for sure, and not just because of dropping pieces. If you think of one program playing both boards, it is also an interesting (and complicated) challenge.

I've thought about it off and on over the years, but never took the plunge...

Bob, you talk about "Bughouse" (2 vs 2), but the topic is the more simple (but already very tricky) Crazyhouse (1 vs 1).

First time I ever played something called crazy house it was 2 vs 2. I assumed they were the same based on that.

What pieces do you "drop" if you have no way to get pieces of the same color from your partner???

I'm going to look it up since I appear to be talking about something different (but a more fun game with a partner anyway).

OK, looked at it. Appears you need two chess sets so that when you capture an opponent's piece, you hang on to it but change the color to your side...

At least an easier game to program without the communication, but it has an absolutely enormous branching factor. One might be a little selective and reduce that but it would be interesting...

[bob]

This might be an interesting project for a retired faculty member.

[Evert]

OK, looked at it. Appears you need two chess sets so that when you capture an opponent's piece, you hang on to it but change the color to your side...

At least an easier game to program without the communication, but it has an absolutely enormous branching factor. One might be a little selective and reduce that but it would be interesting...

You can do some very agressive forward pruning with drops. Stuff that isn't safe, doesn't threaten promotion and doesn't threatens the king is fair game for pruning or (severe) reductions. The way I do this is very crude (because my program is very general), I'm sure one can improve there easily.

[Ferdy]

This might be an interesting project for a retired faculty member.

Perhaps the best human zh players can be found here.

https://en.lichess.org/player

Some claimed they can score around a max of 55% at TC 3'+0, against some of the top best zh engines around.
That TC is optimal for the humans, as they try not to allow the engine to think deeper.
Most engines at this TC could only reach an average of depth 8.

[Ferdy]

OK, looked at it. Appears you need two chess sets so that when you capture an opponent's piece, you hang on to it but change the color to your side...

At least an easier game to program without the communication, but it has an absolutely enormous branching factor. One might be a little selective and reduce that but it would be interesting...

You can do some very agressive forward pruning with drops. Stuff that isn't safe, doesn't threaten promotion and doesn't threatens the king is fair game for pruning or (severe) reductions. The way I do this is very crude (because my program is very general), I'm sure one can improve there easily.

LMR and LMP are also effective as only captures, threat to capture, checks, threat to check, drop check, drop to check, drop to capture and drop to promote matters most of the time.

[gbtami]

This might be an interesting project for a retired faculty member.

Perhaps the best human zh players can be found here.

https://en.lichess.org/player

Some claimed they can score around a max of 55% at TC 3'+0, against some of the top best zh engines around.
That TC is optimal for the humans, as they try not to allow the engine to think deeper.
Most engines at this TC could only reach an average of depth 8.

If someone interested about CH and the best human player:
https://hu.lichess.org/blog/VvFZfiUAAD2 ... use-player