> Edward de Grijs <grey.matter <at> hotmail.com> writes > > Hi Stefan, > thanks for your reply. > Now I understand what you have done, it is a lot of work. > A few remarks: > I find it interesting that you are using such a broad range of programs. > Many of them are not present on CGOS.
Yes here is one of my aims, my goal would be to create an overview of all the programs available or having participated in other tournaments. I was happy, that Remi Coulom did make publicly available his version of CrazyStone that won Gold in Turin 2005 and as soon as my first new results will be published, I really want to ask Eric van der Werf, if I could get a copy of his program steenvreter, which won gold in Amsterdam 2007. Do you have some contact with him or with some other programmers in the Netherlands? > Now you are stating that you have an universal interface by which you > can interface the program directly to any other program. > Did you though about creating an interface from every program to GTP > (if necessary) and running the games against the pool on CGOS? > By rotating the different programs it would be possible to get a rating > for each program on CGOS, which includes the newest programs by > default. This sounds interesting for me. Yes, I have tgought exactly of this and want to do it if I can find the time I need - unfortunately I am not a programmer yet - Peter Woitke did make every programming work I asked him for my tournaments! But he is very busy working on Suzie now. So I want to try these things myself - I am quite sure I would find all the help I would need here or on other forum+mailing lists - so it is really only a question of time ... BtW. Hiroshi Yamashita seems to have realized exactly what we just think of - interesting coincidence! See the games CrazyStone vs. KCC Igo discussed here in the list. > The new Monte Carlo programs are almost all using chinese rules only, > while I have read that you are japanese rules. How do you cope with > this now? Oh yes, I think I still prefer japanese rules even they are often tricky, unclear and not really suited for computers but I like very much computer programs acting like humans would do... But this is not very important as I don´t need fixed rules for my tournament. Just two programs that play against each other have to consent on which rules are applied. So I perhaps would have to avoid a pairing between a program that only plays chinese rules like some MC-programs do and one that only plays japanese rules like some older programs. But if possible I would try to never exlude any program from participation only of cause of the rules. A problem very much more important is the time used! Only 1 year ago, I could let nearly every program play with as much time as it was able to take. GnuGo 3.2 was the first one I could show being very much stronger with more time. In my 13x13 tournament I first tried to let play version 3.2 on level 10 like the version 3.0 had played and the result was nearly the same as with the older version. But then playing on level 15 GnuGo 3.2 was so much stronger that it could take third place in the whole tournament! Today nearly every program has efficient scalability with time, so I have to set a time limit but I don´t want to test things like time-management! My purpose ever was to test the playing strength and nothing else. If a program crashes while playing, the game will be repeated and if possible even continued exactly at the point where it was left. I report bugs and everything that goes wrong but I am not testing functionality and stability. So I think I will set the time as one descripting part of the playing engine comparable to the number of the version of the program. E.g. there could be one engine MoGo(10min/game) and another engine MoGo(30min/game)! And firstGo, how strong do you think it is by now on 9x9 and on 19x19? > Success, > Edward. THANKS! Thanks a lot for every suggestion or comment! Stefan _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
