On Wed, 2008-12-10 at 12:39 -0800, terry mcintyre wrote: > Most of those 55 distinct moves are rarely used in the opening. I once > heard a simple rule which seems to cover just about everything > interesting: "consider only moves which are on the 3rd and 4th lines, > and/or within a manhattan distance of n, for some small n, of some > other stone already on the board." If memory serves, David Fotland > mentioned this at the Portland Congress. Some players favor opening > moves on the fifth line, however.
I'm testing this rule now with the reference bot. I have 10 versions playing now - sets of 5 at 2 different levels, 200 and 2000 playouts. All version have some of my simple enhancements such as the Michael Williams rule. here are the 5 versions: A version without the rule. 2 versions where N = 2 2 versions where N = 3 A neighbor stone must be less than or equal to N distance away if you are not on line 3, 4 or 5. So N = 3 includes knight move patterns. For each N I test, I have a version that does it in the playouts and a version that does it only when making the final choice. If you use this rule in the playouts you really need to also apply it when making your final move choice so my play-out version does both. The 5th line is the e5 point on 9x9 boards. It's interesting that I don't notice a slowdown even when testing for this rule in the playouts. For each intersection on the board I pre-built an array of points to test so that this is mostly table driven. (There can be up to 24 points to test, but obviously this doesn't happen on 9x9 boards since the point in question is near the edges.) When testing, I played a few games between 2 version at 10 playouts and looked carefully at the moves to verify that it was working. The version without the rule would occasionally play a move like A1 even when nothing was near. But this rule on 9x9 doesn't fire all that much once the first few moves have been played. One of the levels I'm testing is very low because I suspect this rule will be more helpful at low levels and not as helpful at higher levels, but that is just a hunch. I'll try to send some results tomorrow. - Don _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
