📝H. G. Muller wrote on Tue, Jan 3, 2023 09:54 PM UTC:
I made a first attempt to make the Diagram play this. At the moment it still doesn't know the rule that you cannot leave the Duck where it is, although it effectively enforces that rule on the opponent, because its moving is triggered by moving the Duck.
This simplification makes it possible to treat it similar to Refusal Chess: the search just keeps track of the two best moves (in absence of the Duck), and only uses the score of the best if that is a contact capture (assuming the Duck would block it otherwise). In Refusal Chess you would always ignore the best move, assuming it will be refused. A refinement is that moves that have the same blocking square as the best move cannot become second-best (because then both could be blocked simultaneously). It only considers blocking on the first step of a slide, so it will overlook 'cross blocking' (where the Duck would block two moves going over the same square in different directions, and these happened to be best and second-best).
It will also overlook the possibility of 'reserving' your next move, as this cannot be done when moving the Duck is not mandatory. This will be hard to implement without making it very complex. Only moves with a single blocking square can be reserved, and it only makes sense to reserve a move that would refute the opponent's best reply to your current move. This would be an alternative method of 'disarming' his best move, which can be used when it is not possible to block that move. If it was blockable, so that both methods are available, it would make no difference which method you used, as both would make him fall back on his second best move. Unless the move you reserve refutes that move too. The chances that several moves are refuted by the same (reservable) move are a lot larger than those that you can block several independent moves (i.e moves that are not part of the same slide): you only need a hefty threat (like a check), and most of his moves are out. It must be possible to design a not-too-complex algorithm based on that idea.
I made a first attempt to make the Diagram play this. At the moment it still doesn't know the rule that you cannot leave the Duck where it is, although it effectively enforces that rule on the opponent, because its moving is triggered by moving the Duck.
This simplification makes it possible to treat it similar to Refusal Chess: the search just keeps track of the two best moves (in absence of the Duck), and only uses the score of the best if that is a contact capture (assuming the Duck would block it otherwise). In Refusal Chess you would always ignore the best move, assuming it will be refused. A refinement is that moves that have the same blocking square as the best move cannot become second-best (because then both could be blocked simultaneously). It only considers blocking on the first step of a slide, so it will overlook 'cross blocking' (where the Duck would block two moves going over the same square in different directions, and these happened to be best and second-best).
It will also overlook the possibility of 'reserving' your next move, as this cannot be done when moving the Duck is not mandatory. This will be hard to implement without making it very complex. Only moves with a single blocking square can be reserved, and it only makes sense to reserve a move that would refute the opponent's best reply to your current move. This would be an alternative method of 'disarming' his best move, which can be used when it is not possible to block that move. If it was blockable, so that both methods are available, it would make no difference which method you used, as both would make him fall back on his second best move. Unless the move you reserve refutes that move too. The chances that several moves are refuted by the same (reservable) move are a lot larger than those that you can block several independent moves (i.e moves that are not part of the same slide): you only need a hefty threat (like a check), and most of his moves are out. It must be possible to design a not-too-complex algorithm based on that idea.