hgm wrote:Weird game against Sjaak II, which prefers delivering a perpetual check over checkmate in one:
I'll investigate. I've been fixing things anyway. I just hope I can reproduce the problem.
Unrelated, but did you see the update to my promotion-input question? No rush if you're busy, but there a,re some things where I'd like to know if I'm running into XBoard limitations, or messing up somehow.
hgm wrote:
The name of the engine is CrazyWa, because eventually it will play all Shogi variants as well, in particular Wa Shogi with drops, a Shogi variant on an 11x11 board. None of the other variants is currently operational, (or at least not tried even once), though.
hgm wrote:I'm not sure if it is worth it, (as I hope to have much better versions in a few days), but I have uploaded the first more-or-less usable version of my engine to my website. It beat Pulsar 22.5-17.5 at 40 moves/min, with only one illegal move (which I adjudicated as a win, because the problem was that a mate-in-one move was not always returned by the search, so that it repeated the previous move. But this is now fixed).
The name of the engine is CrazyWa, because eventually it will play all Shogi variants as well, in particular Wa Shogi with drops, a Shogi variant on an 11x11 board. None of the other variants is currently operational, (or at least not tried even once), though.
This version cannot castle (but accepts castling from the opponent), and it also seems that it doesn't realize e.p. capture can be a method of check evasion. It is completely untuned.
No, HaChu is an engine for dropless games on very large boards. You could say it is a successor of Shokidoki, which so far was my only engine that could handle drops. It was not easy in Shokidoki's board and move representation to implement castlings and e.p.capture, or do boards larger than 9x9 or more than 16 piece types per side. CrazyWa is an experiment with a new board representation, which could do such things, still in a framework of byte-size square encoding. The new board representation can in principle do boards upto 12x11 with 64 piece types per side (of which maximally 54 in hand). It can also do promotion on drops (as in Kyoto Shogi), as drops are encoded per from-square, and promotions per to-square, using the off-board squares. (For Chess the board is only 8x8,so there are plenty of extra off-board squares on 9th and 10th rank to encode the special moves.) If I am happy with it, I might switch Shokidoki to that same board & move representation, effectively merging CrazyWa and Shokidoki.
I did notice your XBoard question, but I didn't have time to look into it yet. The original question (about the stationary promotion) was also not easily answered, and it seems I have to dig deepinto the code to understand what is going on. Now that CrazyWa works I will have more time to do that, while test games are in progress.
Indeed, drops are a very good way to prevent draws. There never is a natural end to a game, and 3-fold repetitions are really the only cause. But it is rare that both engines would allow that. In Shogi there still is the possibility of an 'impasse', when both Kings get to the promotion zone, because Shogi pieces are asymmetrically forward directed. But I don't think you can have that in Crazyhouse.
Evert wrote:
I'll investigate. I've been fixing things anyway. I just hope I can reproduce the problem.
Ok, found it.
The problem is the special mate-search. By default, it's activated for drop games (it can be disabled in the options dialog) but it doesn't test for repetitions. For some reason it prefers the longer mate, not realising it's walking into a repetition (or indeed, keeps preferring round-about mates to direct mates).
It's easy to work around, but a bit harder to fix properly.
