I haven't been using CGOS at all lately (planned gap in development around the holidays). When I do start up again, it'll probably be on the 9x9 server. I'd do this because of the following reasons: * Games finish faster on the 9x9 server * Current testing/tuning is done on 9x9 (I have a short attention span while playing a weak bot) * Strength differences are tougher to overcome on large boards To overcome this, a handicap system or a denser field would be required * Replaying a game while debugging is shorter on 9x9 * There are more opponents on the 9x9 server * Picking only one server to run on is easier than doing both (affects setup, baby sitting, and hardware consumed)
I think that many of those reasons are more relevant for weaker bots. I suspect that the 19x19 server will be less popular with weaker bots for those reasons. I could put up a non-searching bot that does ok on 19x19, but I'm currently hesitant to do that. In the past, I've put up bots on 9x9, but they required baby sitting because the connection to the server was not maintained. After many restarts, I eventually got lazy and nobody complained... On Jan 8, 2008 2:24 PM, David Fotland <[EMAIL PROTECTED]> wrote: > I think there are two reasons there are not more programs on 19x19 CGOS: > > 1) The anchor, Gnugo, is quite strong, Many Faces 12 is stronger, and > CrazyStone is much stronger. Since the programs playing are so strong, it > is demoralizing for a new program to lose so often. Without weaker > competition, it is hard to get accurate ratings for new, weaker programs. > > 2) The rounds take almost an hour, so it takes much too long to get enough > games to see how your program is doing. In my local testing, I use 10 or > 15 > minutes per side. I like to see 50 to 100 test games get any confidence > that a new version is stronger. I prefer to get a test run complete in > under a day. > > I propose the following changes: > > I'll put up 3 weaker versions of Many Faces 12, so there is competition at > lower ratings. These versions are quite fast, using only a few seconds > for > a full game. This will provide some stable opponents for the weaker > programs. > > I think the time limits for CGOS19x19 should be reduced to 10 or 15 > minutes > per side. This is enough to test programs, and it's still a reasonable > time > limit for games against people. Since programs that search scale with > additional time, the relative ratings of these programs should be similar > at > 10 minutes per game and 30 minutes per game. > > -David > > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ >
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
