My program only stores 1 canonical position in symmetrical instances. But that is only for the opening book. Where there is a choice of symmetrically equivalent moves, it makes one at random.
For instance, there might only be one canonical response to e5 in the opening position but this could translate to 4 or 8 actual moves and one of these would be randomly selected. I concluded that there would be little benefit and a lot of trouble trying to take advantage of symmetries in the search. The very first move would benefit enormously, but after that the benefit is sharply reduced to almost nothing after another move or two. At least in positions likely to occur in games. - Don On Tue, 2007-06-19 at 03:52 +0200, Magnus Persson wrote: > > > > But I would still like to know how many MC evaluations it would take > > until all corners look at least somewhat similar... I bet that is many > > more than we see used currently - if it ever gets there. > > The approach I take in Valkyria is radically different. I simply prune all > symmetric moves. For example on 9x9 without a book, Valkyria will only > evaluate > 5 moves 3-3, 3-4, 3-5, 4-4, 4-5, and 5-5. But even doing so the real > difference > between these moves if ther is any is so small it is a waste of time trying to > find it out so when the book is used any of these moves are played with equal > probability. > > I added this feature to increase the opening strength in 7x7, but although it > helps it still just fraction what is necessary to actually find the best line > of play for sure. > > -Magnus > _______________________________________________ > 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/
