I had a look at the way KGS scores games, and I think I'm OK with just using the same rules as on CGOS, right? As long as I don't implement final_status_list and kgs-genmove_cleanup everything should work as expected and I can assume all stones on the board are alive, right?
Urban On Fri, Mar 6, 2015 at 10:25 AM, Urban Hafner <[email protected]> wrote: > Hey everyone, > > I'm still at the early stages of development of my bot > <https://github.com/ujh/iomrascalai>. Right now it only supports > Tromp-Taylor rules as this is what CGOS uses (well, almost). This is a bit > problematic, as for one my bot and GnuGo seldomly agree on a score (even > when running GnuGo 3.8 with "--chinese-rules --positional-super-ko > --capture-all-dead --score aftermath --play-out-aftermath") so running > benchmarks against GnuGo isn't completely useful. The other problem is of > course that Nick uses chinese rules for his KGS tournaments. > > Now my question is what exactly do I have to do to implement chinese > rules? If I understand the differences correctly, the only difference > between TT and Chinese are (I somehow couldn't find any mention of > positional super ko in the TT rules on Senseis > <http://senseis.xmp.net/?LogicalRules>, but IIRC both use PSK): > > 1. Suicide isn't allowed in chinese rules > 2. Dead stones are removed before scoring in chinese > > Now #1 is simple and I've already implemented it as it's used by CGOS, but > how would I handle #2? I guess it needs to be a reasonably fast as it > should be used even in the playouts. It seems like a rather difficult and > error prone approach, or is there a quick way to find groups that are 100% > alive? > > The alternative would be to continue using TT rules internally and just > play until there are only single point eyes left on the board. > > Urban > -- > Blog: http://bettong.net/ > Twitter: https://twitter.com/ujh > Homepage: http://www.urbanhafner.com/ > -- Blog: http://bettong.net/ Twitter: https://twitter.com/ujh Homepage: http://www.urbanhafner.com/
_______________________________________________ Computer-go mailing list [email protected] http://computer-go.org/mailman/listinfo/computer-go
