hgm wrote:
Anyway, glad that you made progress. It is good to see so many new programs, it will make the tourney really exciting.
Absolutely!
In a way, it's a shame that there aren't more resources (even if it's just a website that lists available engines and some of their features) for general variant engines, it seems to be a real niche!
Well, I still have plans to write an installer tool for engines that would download and install those the user selects from the list it presents, based on info it first fetches from a maintainer site. Once I have that, I will probably act as maintainer for variant engines, and set up someweb pages with on-line info for it as well. Sort of a more extensive version of what I have now for Xiangqi.
Some unforeseen difficulties in the interface code, but it all seems to be working correctly now. I just played my first game and managed to get completely slaughtered, which is slightly embarassing since both the evaluation and the SEE function are basically stubs. I'm going to play some more games to check that everything seems to be working, then I'm going to get those bits working (I guess I'll release the new version at that point as well).
Hopefully I can then set my desktop to work over the weekend to start evaluation tuning while I look into producing attack tables to speed up the in-check detection.
Between the nice weather and chasing a number of nasty bugs (including one that must have been there forever) I didn't get round to tuning anything until last night, so I let things run overnight.
Only to find this morning that because I hadn't yet implemented piece-square tables for arbitrary boards the program is completely unable to win any end games, making the tuning test almost completely uesless.
I also need to sit down and actually do something about passed pawn and candidate evaluation...
I also broke compilation in 32-bit mode, but that was at least somewhat deliberate in the sense that I knew that would happen while I focussed on getting things to work before making them portable. This at least will be straightforward to fix again.
Well, I guess that the enhancements I made to Spartacus to prepare it for Leiden will also have broken it for most variants other than normal Chess. As so far they seem to perform without problems, I will generalize them to the other variants.
Ok I think I managed to get Heretic working. It's unfinished, only plays capablanca and very weak (probably still many bugs) and I don't have time to work on it
The download link is here: http://www.mediafire.com/download.php?htw2wp622nfnza4
Please let me know if there are problems with it. Last addition was setboard but I didn't test it thoroughly. I expect it to fight for the last place
I was bored waiting for the test results to come in (and I was on the train), so I implemented analyse mode and pondering in Sjaak today (actually I just ported the relevant code over from Jazz). I guess that's not going to be directly useful for the tournament.
The tuning itself isn't going so well, from the look of things. I haven't actually checked the games yet, but so far the draw-rate is ~67%. This is self-play, so I guess a high draw rate is expected, but even so this seems very high to me (Jazz gets ~30-35% draws in self-play). I suspect Sjaak may still have trouble winning simple end-games, but another problem could be that it's a statistical fluke that is exacerbated by playing very similar games from the starting position. Sjaak does mild randomisation of move ordering at the root at the beginning of the game, but that may not be enough. Are there a set of recommended testing positions?
Hmm, the draw rate in 10x8 Chess should be around 16%.
(For normal Chess I find indeed ~32%.) This is what Fairy-Max and Joker80 have in self-play. What material is typically on the board, when a game ends draw? End-games should not really be different much from standard Chess; the low draw rate is because usually the game is decided long before the end-game, by devastating tactics with the super-pieces.
I am not aware of any test positions, sorry about that.
hgm wrote:Hmm, the draw rate in 10x8 Chess should be around 16%.
(For normal Chess I find indeed ~32%.) This is what Fairy-Max and Joker80 have in self-play. What material is typically on the board, when a game ends draw?
I'll check that.
End-games should not really be different much from standard Chess; the low draw rate is because usually the game is decided long before the end-game, by devastating tactics with the super-pieces.
Makes sense. I have the impression (from looking at a few games) that Sjaak may be mis-managing those as well, which is a symptom of badly balanced evaluation. Don't think that'd cause the high draw-rate though.
I am not aware of any test positions, sorry about that.
Hmm... I'll see if I can generate something suitable. If I throw out repetitions, flank pawn pushes (for simplicity) and non-developing moves a modified perft 6 or 8 should give a nice selection of candidate positions. Filter out everything that has an unbalanced position, hanging pieces or otherwise obvious threats (determined by a quick search) should give something usable to begin with.
