Are there more experimental results coming? Have you tried going higher than 100 yet? How much memory do you need at 100 during the course of a 30 minute game?
On 1/19/07, Don Dailey <[EMAIL PROTECTED]> wrote:
I'm not sure it matters, because as I reported earlier you can delay node expansion with hardly any effect on the strength, and I'm requiring 100 hits before expansion. Pruning out moves from the tree WILL weaken the program - the question is how much? There is a tradeoff of course. What I am doing is pre-allocating the memory, my program does not need to malloc or free it. I start at zero, and have a global counter that tells me which entry to use next. I use 100 hits so I don't run out of entries for quite a while. To "free" this memory I just reset this global pointer to zero. The children of a node are listed in memory sequentially, I don't have to store separate pointer to each of them, I just have a first child pointer (index) and a count of the number of children. So when I open up a new node I find all the legal moves and use the next N available slots to place them. This is wasteful, but it's dirt simple and very fast. Even with only 256 MB of memory I have plenty for a pretty long search. - Don On Fri, 2007-01-19 at 13:09 -0300, Eduardo Sabbatella wrote: > I see your point. > > I was thinking that after X simulations, transposition > table could be made 'persistent'. Not before that. > > So this will avoid a lot of 'garbage', not so 'useful' > board states. > > I have to think a lot about it. But I suposse as my > transposition table implementation, it should have > different node sizes (4 moves, 8, 16 or full board.) > Some precalc data (at least the number of moves in > this state as, as its usefull for UCT). > > How to avoid to recalculate discarted data? Perhaps > its all about choosing wisely what to discard and when > to store permanently. > > > > --- Don Dailey <[EMAIL PROTECTED]> escribió: > > > On Fri, 2007-01-19 at 11:33 +0000, Eduardo > > Sabbatella wrote: > > > If this ratio becomes tooo low (ex, win ratio > > below 1% > > > with 100000 games confidence), you can be sure you > > can > > > delete all the game tree after this move. All the > > > transposition tables following this move. > > > > I think you can win, but please note that the > > weak moves have very small tree's, so you have to > > clean > > up a LOT of them. And you ARE throwing out > > information that UCT will eventually reconstitute. > > > > - Don > > > > > > > > > > _______________________________________________ > > computer-go mailing list > > computer-go@computer-go.org > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > > > > > __________________________________________________ > Preguntá. Respondé. Descubrí. > Todo lo que querías saber, y lo que ni imaginabas, > está en Yahoo! Respuestas (Beta). > ¡Probalo ya! > http://www.yahoo.com.ar/respuestas > > _______________________________________________ > 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/
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/