On Wed, 2007-01-03 at 22:04 -0800, Peter Drake wrote: > How much time should a program spend on each move? > > If my program has t milliseconds left to use in a game, and there are > an estimated m moves left on the board (e.g., this many vacant > spaces), one reasonable choice is t / m.
Excellent question. I think this really is a reasonable choice. I did a LOT of tests to determine this and it makes sense to front load quite heavily. I used a constant, I did not try to estimate how many moves might be left. If you use t/m you really should make m much smaller than the number of vacant points left. It's not a waste to spend a lot of time on early moves. From casual observation, I noticed that most games were decided very quickly, after just a few moves in 9x9. Another reason to front load is that the game gets easier and easier to play correctly as more stones get placed. It's a matter of concentrating the most energy where it's needed. My program notices when the game is pretty much a forgone conclusion and when this happens it plays even faster - I do this so that I can be even more aggressive about earlier moves. > In practice, this seems to spend too much time on early moves, which > (under UCT/MC) is largely wasted time. Would it be better to use > something like t / m**k, for some constant k? (Looking at graphs of > such functions, k = 1.5 seems reasonable.) You should test all of this. That's what I do. I think self-testing of different formula's and constants is fine for this kind of thing. > It would also be interesting to look at the graphs of how much time > humans spend on each move; is it usually less for the opening moves > than for middle / endgame moves? Is there a smooth curve, or is there > a relatively abrupt shift from joseki to analysis? > > Peter Drake > Assistant Professor of Computer Science > Lewis & Clark College > http://www.lclark.edu/~drake/ > > > _______________________________________________ > computer-go mailing list > [email protected] > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
