Comments/Ratings for a Single Item

I'd like to point out that I've made a jocly implementation of Seireigi shogi for my own use and also to better understand the "drop model" made by HG Muller in his shogi for jocly.

I hope it conforms to the rules. Note, however, that there's undoubtedly still a lot of room for improvement, particularly as regards kanji graphics.

The Jocly implementation has two mistakes in it. Please fix these.

1. Player B's Heavenly Horse (promoted Knight) has its jumping moves reversed (the stepping moves are fine).
2. Pieces that get dropped into the promotion zone are able to promote, which shouldn't be possible.

As for the Kanji, the current implementation works fine, but may be a problem if you want to implement the larger Seireigi variants in Jocly as well.

Otherwise, stellar implementation.

Player B's Heavenly Horse move should be fixed now.

Promotion improvements still to be made...

Now the pieces that get dropped into the promotion zone should no longer be able to promote.

Thank you.

I was thinking about porting your Jocly implementation of Seireigi over to this site. I looked at your site's Jocly files and saw a bunch of files beginning with "seireigi-shogi" in the directory "https://biscandine.fr/variantes/shogi/dist/browser/games/chessbase/". Are these the only files that I need if I want to port the implementation, or is there something else that I am missing?

Edit: I have added a link to the implementation on your site for now.

For integration on the chessvariant site, it's probably best to wait until the work on integrating shogis into jocly has progressed. HG Muller will be able to do it easily, but it may be a bit early for us since it might change.
Otherwise, for hosting on a personnal site, the following zip contains everything you need

It could be further improved by setting a relative value to the pieces.

To make a new Jocly game available on the site here, you would have to upload its *-model.js, *-view.js and *-config.js in the game library (the dist/browser/games/chessbase that is created after running gulp for building) to /play/jocly/dist/browser/games/chessbase/ on CVP. The model and view files from the source code won't do; the building process concatenates these with the base-model and view, and other sub-models that are specified in index.js.

You then also have to add a line for the game in /play/jocly/dist/browser/jocly-allgames.js to make it appear in the list of 'other Jocly games'. If you want it to be displayed together with a thumbnail there, you should also upload that to where the allgames line says it is. If the game uses dedicated resources, such as new pieces, these files would have to be uploaded too.

The whole effort of this will only become visible if both your browser's cache and the CloudFlare cache have flushed out the old allgames file. You can do the first thing yourself, but the normal way of pressing Shift while loading the Jocly page doesn't work; you have to explicitly view the allgames file yourself, and reload that without caching. And over CloudFlare users have no control at all. You'd have to wait a few weeks, or ask Fergus.

I also have a draft of kyoto shogi available on my home page.

I don't yet know whether it complies perfectly with the rules. It's also based on the shogi model currently being integrated into jocly.

The Kyoto Shogi model looks good, except for one mistake. There are no restrictions on drops in Kyoto Shogi, except that the drop square must be empty.

I've lifted the restrictions on lance and horse. It should be better now.

To play reasonably, Kyoto Shogi would probably need dedicated evaluation. Piece values are not the dominant evaluation term, not even if you average those of the alternating types. Most important is (future) mobility, as some pieces have irreversible moves, and are doomed to get stuck. Fourth-rank Knights are worse than useless; they form an indestructable shelter for the opponent's King, and force you to dedicate a piece for protecting them from capture by that King. Having a useless piece, or even a piece that is a liability, is still better than giving the opponent an extra piece in hand...

At the moment the Kyoto Shogi is in reserve on my github shogivars branch.

I don't think I'll be able to add the evaluation function on my own. If you have the time, the inclination and if you think this variant could be of interest to jocly, help would be welcome to finalize it.

By the way is there a Pawn Limit > 1 on a column for this variant?

The Kyoto Jocly implementation is definitely better.

There are no restrictions on drops in Kyoto Shogi, other than that the square must be empty. (So no, there is no one Pawn per column rule in this game.)

Fortunately, all you need to do to fix this is force captured Knights to change to Golds, force Lances to change to Tokins, and force Pawns to change to Rooks.

Unfortunately, I have no experience programming games in Jocly, so I am unable to add an evaluation function.

There's everything in the model eventually to set the number of pawn per column.

The rule should be OK now. Thanks for the tests.

Regarding the evaluation function, the suggestion was for HGM, which seems to have a great deal of experience in programming shogis.

Yep, I did implement Kyoto Shogi in my engine CrazyWa. (Which is a multivariant engine especially for games with piece drops.) When I have the opportunity I will copy its piece-square tables for use in the Jocly implementation.

Speaking of rules, I took a look at your rules for Seireigi, and noticed that the image for the Great Elephant's move diagram is not lined up with with the piece's actual move. The current image shows a (disconnected) dot two spaces ahead of the piece, and no backwards diagonal step. There should be a dot on all spaces adjacent spaces except the one directly behind the piece, along with the two ranging moves.

The Jocly implementation has the correct move.

This has now been corrected.

Thanks for your vigilance.

The author, A. M. DeWitt, has updated this page.

Note that the Sjaak II engine is configurable for drop variants. It could be that Nebiyu is too (as I know it to support mini-Shogi). CrazyWa is not configurable, but it is open source, (a C program), and can probably easily be adapted to play Seireigi: it is an engine designed for playing drop variants, and already supports Wa Shogi.

Since Fairy-Stockfish supports regular Shogi, I suppose it can be configured for other drop variants too. (Caveat: although Fairy-Stockfish is in general super-strong, it derives its strength in Shogi mainly from a trained neural network evaluating the positions. I heard that this increased its strength by about 1000 Elo over using the hand-crafted evaluation.)

Hello HGM,

"I heard that this increased its strength by about 1000 Elo over using the hand-crafted evaluation"

Have not trained a NNUE for Sereigi yet, but indeed, in Shogi the NNUE was over 1000 Elo above HCE. For now, I have defined Sereigi for FSF and integrated it in the Winboard GUI, just download it (https://drive.google.com/file/d/1STneJS83cyom0qGzytWgDwtemdc2WE2F/view?usp=sharing), unzip it and start it in Winboard (it's the only engine I have included in the factory.ini file).

With the graphics via pieceToCharTable and the bitmaps folder of Winboard there are problems (could only apply the bmp of archbishop for the lance piece) Of course I can change the bmp's in the bitmaps folder, but that would make the Winboard GUI unusable for other games. The best for now may be to start manually from the command line

FSF2304 load variants.ini

xboard

memory 2048

cores 22 (the number of cores your processor has)

variant sereigi

st 10 (seconds per move)

and type some move in the notation g1f2 If a piece promotes, the sign + after the move is necessary, for example h8b2+

I would say, even without NNUE it plays at superhuman strength and is without doubt atm world's strongest Sereigi player, Enjoy

I'm seeing the Tokin having the same image as the Pawn in the interactive diagram

Not sure how that is possible since the diagram clearly defines the Tokin has having the p2 image instead of the p image.

Maybe refresh your browser cache?

I tried to load the Seireigi engine you put a link to, but I wasn't able to get it to work. It said it couldn't find the variants.ini file upon loading.